Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1 | #!/usr/bin/env perl |
| 2 | # |
| 3 | # Implemented as a Perl wrapper as we want to support several different |
| 4 | # architectures with single file. We pick up the target based on the |
| 5 | # file name we are asked to generate. |
| 6 | # |
| 7 | # It should be noted though that this perl code is nothing like |
| 8 | # <openssl>/crypto/perlasm/x86*. In this case perl is used pretty much |
| 9 | # as pre-processor to cover for platform differences in name decoration, |
| 10 | # linker tables, 32-/64-bit instruction sets... |
| 11 | # |
| 12 | # As you might know there're several PowerPC ABI in use. Most notably |
| 13 | # Linux and AIX use different 32-bit ABIs. Good news are that these ABIs |
| 14 | # are similar enough to implement leaf(!) functions, which would be ABI |
| 15 | # neutral. And that's what you find here: ABI neutral leaf functions. |
| 16 | # In case you wonder what that is... |
| 17 | # |
| 18 | # AIX performance |
| 19 | # |
| 20 | # MEASUREMENTS WITH cc ON a 200 MhZ PowerPC 604e. |
| 21 | # |
| 22 | # The following is the performance of 32-bit compiler |
| 23 | # generated code: |
| 24 | # |
| 25 | # OpenSSL 0.9.6c 21 dec 2001 |
| 26 | # built on: Tue Jun 11 11:06:51 EDT 2002 |
| 27 | # options:bn(64,32) ... |
| 28 | #compiler: cc -DTHREADS -DAIX -DB_ENDIAN -DBN_LLONG -O3 |
| 29 | # sign verify sign/s verify/s |
| 30 | #rsa 512 bits 0.0098s 0.0009s 102.0 1170.6 |
| 31 | #rsa 1024 bits 0.0507s 0.0026s 19.7 387.5 |
| 32 | #rsa 2048 bits 0.3036s 0.0085s 3.3 117.1 |
| 33 | #rsa 4096 bits 2.0040s 0.0299s 0.5 33.4 |
| 34 | #dsa 512 bits 0.0087s 0.0106s 114.3 94.5 |
| 35 | #dsa 1024 bits 0.0256s 0.0313s 39.0 32.0 |
| 36 | # |
| 37 | # Same bechmark with this assembler code: |
| 38 | # |
| 39 | #rsa 512 bits 0.0056s 0.0005s 178.6 2049.2 |
| 40 | #rsa 1024 bits 0.0283s 0.0015s 35.3 674.1 |
| 41 | #rsa 2048 bits 0.1744s 0.0050s 5.7 201.2 |
| 42 | #rsa 4096 bits 1.1644s 0.0179s 0.9 55.7 |
| 43 | #dsa 512 bits 0.0052s 0.0062s 191.6 162.0 |
| 44 | #dsa 1024 bits 0.0149s 0.0180s 67.0 55.5 |
| 45 | # |
| 46 | # Number of operations increases by at almost 75% |
| 47 | # |
| 48 | # Here are performance numbers for 64-bit compiler |
| 49 | # generated code: |
| 50 | # |
| 51 | # OpenSSL 0.9.6g [engine] 9 Aug 2002 |
| 52 | # built on: Fri Apr 18 16:59:20 EDT 2003 |
| 53 | # options:bn(64,64) ... |
| 54 | # compiler: cc -DTHREADS -D_REENTRANT -q64 -DB_ENDIAN -O3 |
| 55 | # sign verify sign/s verify/s |
| 56 | #rsa 512 bits 0.0028s 0.0003s 357.1 3844.4 |
| 57 | #rsa 1024 bits 0.0148s 0.0008s 67.5 1239.7 |
| 58 | #rsa 2048 bits 0.0963s 0.0028s 10.4 353.0 |
| 59 | #rsa 4096 bits 0.6538s 0.0102s 1.5 98.1 |
| 60 | #dsa 512 bits 0.0026s 0.0032s 382.5 313.7 |
| 61 | #dsa 1024 bits 0.0081s 0.0099s 122.8 100.6 |
| 62 | # |
| 63 | # Same benchmark with this assembler code: |
| 64 | # |
| 65 | #rsa 512 bits 0.0020s 0.0002s 510.4 6273.7 |
| 66 | #rsa 1024 bits 0.0088s 0.0005s 114.1 2128.3 |
| 67 | #rsa 2048 bits 0.0540s 0.0016s 18.5 622.5 |
| 68 | #rsa 4096 bits 0.3700s 0.0058s 2.7 171.0 |
| 69 | #dsa 512 bits 0.0016s 0.0020s 610.7 507.1 |
| 70 | #dsa 1024 bits 0.0047s 0.0058s 212.5 173.2 |
| 71 | # |
| 72 | # Again, performance increases by at about 75% |
| 73 | # |
| 74 | # Mac OS X, Apple G5 1.8GHz (Note this is 32 bit code) |
| 75 | # OpenSSL 0.9.7c 30 Sep 2003 |
| 76 | # |
| 77 | # Original code. |
| 78 | # |
| 79 | #rsa 512 bits 0.0011s 0.0001s 906.1 11012.5 |
| 80 | #rsa 1024 bits 0.0060s 0.0003s 166.6 3363.1 |
| 81 | #rsa 2048 bits 0.0370s 0.0010s 27.1 982.4 |
| 82 | #rsa 4096 bits 0.2426s 0.0036s 4.1 280.4 |
| 83 | #dsa 512 bits 0.0010s 0.0012s 1038.1 841.5 |
| 84 | #dsa 1024 bits 0.0030s 0.0037s 329.6 269.7 |
| 85 | #dsa 2048 bits 0.0101s 0.0127s 98.9 78.6 |
| 86 | # |
| 87 | # Same benchmark with this assembler code: |
| 88 | # |
| 89 | #rsa 512 bits 0.0007s 0.0001s 1416.2 16645.9 |
| 90 | #rsa 1024 bits 0.0036s 0.0002s 274.4 5380.6 |
| 91 | #rsa 2048 bits 0.0222s 0.0006s 45.1 1589.5 |
| 92 | #rsa 4096 bits 0.1469s 0.0022s 6.8 449.6 |
| 93 | #dsa 512 bits 0.0006s 0.0007s 1664.2 1376.2 |
| 94 | #dsa 1024 bits 0.0018s 0.0023s 545.0 442.2 |
| 95 | #dsa 2048 bits 0.0061s 0.0075s 163.5 132.8 |
| 96 | # |
| 97 | # Performance increase of ~60% |
| 98 | # |
| 99 | # If you have comments or suggestions to improve code send |
| 100 | # me a note at schari@us.ibm.com |
| 101 | # |
| 102 | |
| 103 | $flavour = shift; |
| 104 | |
| 105 | if ($flavour =~ /32/) { |
| 106 | $BITS= 32; |
| 107 | $BNSZ= $BITS/8; |
| 108 | $ISA= "\"ppc\""; |
| 109 | |
| 110 | $LD= "lwz"; # load |
| 111 | $LDU= "lwzu"; # load and update |
| 112 | $ST= "stw"; # store |
| 113 | $STU= "stwu"; # store and update |
| 114 | $UMULL= "mullw"; # unsigned multiply low |
| 115 | $UMULH= "mulhwu"; # unsigned multiply high |
| 116 | $UDIV= "divwu"; # unsigned divide |
| 117 | $UCMPI= "cmplwi"; # unsigned compare with immediate |
| 118 | $UCMP= "cmplw"; # unsigned compare |
| 119 | $CNTLZ= "cntlzw"; # count leading zeros |
| 120 | $SHL= "slw"; # shift left |
| 121 | $SHR= "srw"; # unsigned shift right |
| 122 | $SHRI= "srwi"; # unsigned shift right by immediate |
| 123 | $SHLI= "slwi"; # shift left by immediate |
| 124 | $CLRU= "clrlwi"; # clear upper bits |
| 125 | $INSR= "insrwi"; # insert right |
| 126 | $ROTL= "rotlwi"; # rotate left by immediate |
| 127 | $TR= "tw"; # conditional trap |
| 128 | } elsif ($flavour =~ /64/) { |
| 129 | $BITS= 64; |
| 130 | $BNSZ= $BITS/8; |
| 131 | $ISA= "\"ppc64\""; |
| 132 | |
| 133 | # same as above, but 64-bit mnemonics... |
| 134 | $LD= "ld"; # load |
| 135 | $LDU= "ldu"; # load and update |
| 136 | $ST= "std"; # store |
| 137 | $STU= "stdu"; # store and update |
| 138 | $UMULL= "mulld"; # unsigned multiply low |
| 139 | $UMULH= "mulhdu"; # unsigned multiply high |
| 140 | $UDIV= "divdu"; # unsigned divide |
| 141 | $UCMPI= "cmpldi"; # unsigned compare with immediate |
| 142 | $UCMP= "cmpld"; # unsigned compare |
| 143 | $CNTLZ= "cntlzd"; # count leading zeros |
| 144 | $SHL= "sld"; # shift left |
| 145 | $SHR= "srd"; # unsigned shift right |
| 146 | $SHRI= "srdi"; # unsigned shift right by immediate |
| 147 | $SHLI= "sldi"; # shift left by immediate |
| 148 | $CLRU= "clrldi"; # clear upper bits |
| 149 | $INSR= "insrdi"; # insert right |
| 150 | $ROTL= "rotldi"; # rotate left by immediate |
| 151 | $TR= "td"; # conditional trap |
| 152 | } else { die "nonsense $flavour"; } |
| 153 | |
| 154 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 155 | ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or |
| 156 | ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or |
| 157 | die "can't locate ppc-xlate.pl"; |
| 158 | |
| 159 | open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!"; |
| 160 | |
| 161 | $data=<<EOF; |
| 162 | #-------------------------------------------------------------------- |
| 163 | # |
| 164 | # |
| 165 | # |
| 166 | # |
| 167 | # File: ppc32.s |
| 168 | # |
| 169 | # Created by: Suresh Chari |
| 170 | # IBM Thomas J. Watson Research Library |
| 171 | # Hawthorne, NY |
| 172 | # |
| 173 | # |
| 174 | # Description: Optimized assembly routines for OpenSSL crypto |
| 175 | # on the 32 bitPowerPC platform. |
| 176 | # |
| 177 | # |
| 178 | # Version History |
| 179 | # |
| 180 | # 2. Fixed bn_add,bn_sub and bn_div_words, added comments, |
| 181 | # cleaned up code. Also made a single version which can |
| 182 | # be used for both the AIX and Linux compilers. See NOTE |
| 183 | # below. |
| 184 | # 12/05/03 Suresh Chari |
| 185 | # (with lots of help from) Andy Polyakov |
| 186 | ## |
| 187 | # 1. Initial version 10/20/02 Suresh Chari |
| 188 | # |
| 189 | # |
| 190 | # The following file works for the xlc,cc |
| 191 | # and gcc compilers. |
| 192 | # |
| 193 | # NOTE: To get the file to link correctly with the gcc compiler |
| 194 | # you have to change the names of the routines and remove |
| 195 | # the first .(dot) character. This should automatically |
| 196 | # be done in the build process. |
| 197 | # |
| 198 | # Hand optimized assembly code for the following routines |
| 199 | # |
| 200 | # bn_sqr_comba4 |
| 201 | # bn_sqr_comba8 |
| 202 | # bn_mul_comba4 |
| 203 | # bn_mul_comba8 |
| 204 | # bn_sub_words |
| 205 | # bn_add_words |
| 206 | # bn_div_words |
| 207 | # bn_sqr_words |
| 208 | # bn_mul_words |
| 209 | # bn_mul_add_words |
| 210 | # |
| 211 | # NOTE: It is possible to optimize this code more for |
| 212 | # specific PowerPC or Power architectures. On the Northstar |
| 213 | # architecture the optimizations in this file do |
| 214 | # NOT provide much improvement. |
| 215 | # |
| 216 | # If you have comments or suggestions to improve code send |
| 217 | # me a note at schari\@us.ibm.com |
| 218 | # |
| 219 | #-------------------------------------------------------------------------- |
| 220 | # |
| 221 | # Defines to be used in the assembly code. |
| 222 | # |
| 223 | #.set r0,0 # we use it as storage for value of 0 |
| 224 | #.set SP,1 # preserved |
| 225 | #.set RTOC,2 # preserved |
| 226 | #.set r3,3 # 1st argument/return value |
| 227 | #.set r4,4 # 2nd argument/volatile register |
| 228 | #.set r5,5 # 3rd argument/volatile register |
| 229 | #.set r6,6 # ... |
| 230 | #.set r7,7 |
| 231 | #.set r8,8 |
| 232 | #.set r9,9 |
| 233 | #.set r10,10 |
| 234 | #.set r11,11 |
| 235 | #.set r12,12 |
| 236 | #.set r13,13 # not used, nor any other "below" it... |
| 237 | |
| 238 | # Declare function names to be global |
| 239 | # NOTE: For gcc these names MUST be changed to remove |
| 240 | # the first . i.e. for example change ".bn_sqr_comba4" |
| 241 | # to "bn_sqr_comba4". This should be automatically done |
| 242 | # in the build. |
| 243 | |
| 244 | .globl .bn_sqr_comba4 |
| 245 | .globl .bn_sqr_comba8 |
| 246 | .globl .bn_mul_comba4 |
| 247 | .globl .bn_mul_comba8 |
| 248 | .globl .bn_sub_words |
| 249 | .globl .bn_add_words |
| 250 | .globl .bn_div_words |
| 251 | .globl .bn_sqr_words |
| 252 | .globl .bn_mul_words |
| 253 | .globl .bn_mul_add_words |
| 254 | |
| 255 | # .text section |
| 256 | |
| 257 | .machine "any" |
| 258 | |
| 259 | # |
| 260 | # NOTE: The following label name should be changed to |
| 261 | # "bn_sqr_comba4" i.e. remove the first dot |
| 262 | # for the gcc compiler. This should be automatically |
| 263 | # done in the build |
| 264 | # |
| 265 | |
| 266 | .align 4 |
| 267 | .bn_sqr_comba4: |
| 268 | # |
| 269 | # Optimized version of bn_sqr_comba4. |
| 270 | # |
| 271 | # void bn_sqr_comba4(BN_ULONG *r, BN_ULONG *a) |
| 272 | # r3 contains r |
| 273 | # r4 contains a |
| 274 | # |
| 275 | # Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows: |
| 276 | # |
| 277 | # r5,r6 are the two BN_ULONGs being multiplied. |
| 278 | # r7,r8 are the results of the 32x32 giving 64 bit multiply. |
| 279 | # r9,r10, r11 are the equivalents of c1,c2, c3. |
| 280 | # Here's the assembly |
| 281 | # |
| 282 | # |
| 283 | xor r0,r0,r0 # set r0 = 0. Used in the addze |
| 284 | # instructions below |
| 285 | |
| 286 | #sqr_add_c(a,0,c1,c2,c3) |
| 287 | $LD r5,`0*$BNSZ`(r4) |
| 288 | $UMULL r9,r5,r5 |
| 289 | $UMULH r10,r5,r5 #in first iteration. No need |
| 290 | #to add since c1=c2=c3=0. |
| 291 | # Note c3(r11) is NOT set to 0 |
| 292 | # but will be. |
| 293 | |
| 294 | $ST r9,`0*$BNSZ`(r3) # r[0]=c1; |
| 295 | # sqr_add_c2(a,1,0,c2,c3,c1); |
| 296 | $LD r6,`1*$BNSZ`(r4) |
| 297 | $UMULL r7,r5,r6 |
| 298 | $UMULH r8,r5,r6 |
| 299 | |
| 300 | addc r7,r7,r7 # compute (r7,r8)=2*(r7,r8) |
| 301 | adde r8,r8,r8 |
| 302 | addze r9,r0 # catch carry if any. |
| 303 | # r9= r0(=0) and carry |
| 304 | |
| 305 | addc r10,r7,r10 # now add to temp result. |
| 306 | addze r11,r8 # r8 added to r11 which is 0 |
| 307 | addze r9,r9 |
| 308 | |
| 309 | $ST r10,`1*$BNSZ`(r3) #r[1]=c2; |
| 310 | #sqr_add_c(a,1,c3,c1,c2) |
| 311 | $UMULL r7,r6,r6 |
| 312 | $UMULH r8,r6,r6 |
| 313 | addc r11,r7,r11 |
| 314 | adde r9,r8,r9 |
| 315 | addze r10,r0 |
| 316 | #sqr_add_c2(a,2,0,c3,c1,c2) |
| 317 | $LD r6,`2*$BNSZ`(r4) |
| 318 | $UMULL r7,r5,r6 |
| 319 | $UMULH r8,r5,r6 |
| 320 | |
| 321 | addc r7,r7,r7 |
| 322 | adde r8,r8,r8 |
| 323 | addze r10,r10 |
| 324 | |
| 325 | addc r11,r7,r11 |
| 326 | adde r9,r8,r9 |
| 327 | addze r10,r10 |
| 328 | $ST r11,`2*$BNSZ`(r3) #r[2]=c3 |
| 329 | #sqr_add_c2(a,3,0,c1,c2,c3); |
| 330 | $LD r6,`3*$BNSZ`(r4) |
| 331 | $UMULL r7,r5,r6 |
| 332 | $UMULH r8,r5,r6 |
| 333 | addc r7,r7,r7 |
| 334 | adde r8,r8,r8 |
| 335 | addze r11,r0 |
| 336 | |
| 337 | addc r9,r7,r9 |
| 338 | adde r10,r8,r10 |
| 339 | addze r11,r11 |
| 340 | #sqr_add_c2(a,2,1,c1,c2,c3); |
| 341 | $LD r5,`1*$BNSZ`(r4) |
| 342 | $LD r6,`2*$BNSZ`(r4) |
| 343 | $UMULL r7,r5,r6 |
| 344 | $UMULH r8,r5,r6 |
| 345 | |
| 346 | addc r7,r7,r7 |
| 347 | adde r8,r8,r8 |
| 348 | addze r11,r11 |
| 349 | addc r9,r7,r9 |
| 350 | adde r10,r8,r10 |
| 351 | addze r11,r11 |
| 352 | $ST r9,`3*$BNSZ`(r3) #r[3]=c1 |
| 353 | #sqr_add_c(a,2,c2,c3,c1); |
| 354 | $UMULL r7,r6,r6 |
| 355 | $UMULH r8,r6,r6 |
| 356 | addc r10,r7,r10 |
| 357 | adde r11,r8,r11 |
| 358 | addze r9,r0 |
| 359 | #sqr_add_c2(a,3,1,c2,c3,c1); |
| 360 | $LD r6,`3*$BNSZ`(r4) |
| 361 | $UMULL r7,r5,r6 |
| 362 | $UMULH r8,r5,r6 |
| 363 | addc r7,r7,r7 |
| 364 | adde r8,r8,r8 |
| 365 | addze r9,r9 |
| 366 | |
| 367 | addc r10,r7,r10 |
| 368 | adde r11,r8,r11 |
| 369 | addze r9,r9 |
| 370 | $ST r10,`4*$BNSZ`(r3) #r[4]=c2 |
| 371 | #sqr_add_c2(a,3,2,c3,c1,c2); |
| 372 | $LD r5,`2*$BNSZ`(r4) |
| 373 | $UMULL r7,r5,r6 |
| 374 | $UMULH r8,r5,r6 |
| 375 | addc r7,r7,r7 |
| 376 | adde r8,r8,r8 |
| 377 | addze r10,r0 |
| 378 | |
| 379 | addc r11,r7,r11 |
| 380 | adde r9,r8,r9 |
| 381 | addze r10,r10 |
| 382 | $ST r11,`5*$BNSZ`(r3) #r[5] = c3 |
| 383 | #sqr_add_c(a,3,c1,c2,c3); |
| 384 | $UMULL r7,r6,r6 |
| 385 | $UMULH r8,r6,r6 |
| 386 | addc r9,r7,r9 |
| 387 | adde r10,r8,r10 |
| 388 | |
| 389 | $ST r9,`6*$BNSZ`(r3) #r[6]=c1 |
| 390 | $ST r10,`7*$BNSZ`(r3) #r[7]=c2 |
| 391 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 392 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 393 | |
| 394 | # |
| 395 | # NOTE: The following label name should be changed to |
| 396 | # "bn_sqr_comba8" i.e. remove the first dot |
| 397 | # for the gcc compiler. This should be automatically |
| 398 | # done in the build |
| 399 | # |
| 400 | |
| 401 | .align 4 |
| 402 | .bn_sqr_comba8: |
| 403 | # |
| 404 | # This is an optimized version of the bn_sqr_comba8 routine. |
| 405 | # Tightly uses the adde instruction |
| 406 | # |
| 407 | # |
| 408 | # void bn_sqr_comba8(BN_ULONG *r, BN_ULONG *a) |
| 409 | # r3 contains r |
| 410 | # r4 contains a |
| 411 | # |
| 412 | # Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows: |
| 413 | # |
| 414 | # r5,r6 are the two BN_ULONGs being multiplied. |
| 415 | # r7,r8 are the results of the 32x32 giving 64 bit multiply. |
| 416 | # r9,r10, r11 are the equivalents of c1,c2, c3. |
| 417 | # |
| 418 | # Possible optimization of loading all 8 longs of a into registers |
| 419 | # doesnt provide any speedup |
| 420 | # |
| 421 | |
| 422 | xor r0,r0,r0 #set r0 = 0.Used in addze |
| 423 | #instructions below. |
| 424 | |
| 425 | #sqr_add_c(a,0,c1,c2,c3); |
| 426 | $LD r5,`0*$BNSZ`(r4) |
| 427 | $UMULL r9,r5,r5 #1st iteration: no carries. |
| 428 | $UMULH r10,r5,r5 |
| 429 | $ST r9,`0*$BNSZ`(r3) # r[0]=c1; |
| 430 | #sqr_add_c2(a,1,0,c2,c3,c1); |
| 431 | $LD r6,`1*$BNSZ`(r4) |
| 432 | $UMULL r7,r5,r6 |
| 433 | $UMULH r8,r5,r6 |
| 434 | |
| 435 | addc r10,r7,r10 #add the two register number |
| 436 | adde r11,r8,r0 # (r8,r7) to the three register |
| 437 | addze r9,r0 # number (r9,r11,r10).NOTE:r0=0 |
| 438 | |
| 439 | addc r10,r7,r10 #add the two register number |
| 440 | adde r11,r8,r11 # (r8,r7) to the three register |
| 441 | addze r9,r9 # number (r9,r11,r10). |
| 442 | |
| 443 | $ST r10,`1*$BNSZ`(r3) # r[1]=c2 |
| 444 | |
| 445 | #sqr_add_c(a,1,c3,c1,c2); |
| 446 | $UMULL r7,r6,r6 |
| 447 | $UMULH r8,r6,r6 |
| 448 | addc r11,r7,r11 |
| 449 | adde r9,r8,r9 |
| 450 | addze r10,r0 |
| 451 | #sqr_add_c2(a,2,0,c3,c1,c2); |
| 452 | $LD r6,`2*$BNSZ`(r4) |
| 453 | $UMULL r7,r5,r6 |
| 454 | $UMULH r8,r5,r6 |
| 455 | |
| 456 | addc r11,r7,r11 |
| 457 | adde r9,r8,r9 |
| 458 | addze r10,r10 |
| 459 | |
| 460 | addc r11,r7,r11 |
| 461 | adde r9,r8,r9 |
| 462 | addze r10,r10 |
| 463 | |
| 464 | $ST r11,`2*$BNSZ`(r3) #r[2]=c3 |
| 465 | #sqr_add_c2(a,3,0,c1,c2,c3); |
| 466 | $LD r6,`3*$BNSZ`(r4) #r6 = a[3]. r5 is already a[0]. |
| 467 | $UMULL r7,r5,r6 |
| 468 | $UMULH r8,r5,r6 |
| 469 | |
| 470 | addc r9,r7,r9 |
| 471 | adde r10,r8,r10 |
| 472 | addze r11,r0 |
| 473 | |
| 474 | addc r9,r7,r9 |
| 475 | adde r10,r8,r10 |
| 476 | addze r11,r11 |
| 477 | #sqr_add_c2(a,2,1,c1,c2,c3); |
| 478 | $LD r5,`1*$BNSZ`(r4) |
| 479 | $LD r6,`2*$BNSZ`(r4) |
| 480 | $UMULL r7,r5,r6 |
| 481 | $UMULH r8,r5,r6 |
| 482 | |
| 483 | addc r9,r7,r9 |
| 484 | adde r10,r8,r10 |
| 485 | addze r11,r11 |
| 486 | |
| 487 | addc r9,r7,r9 |
| 488 | adde r10,r8,r10 |
| 489 | addze r11,r11 |
| 490 | |
| 491 | $ST r9,`3*$BNSZ`(r3) #r[3]=c1; |
| 492 | #sqr_add_c(a,2,c2,c3,c1); |
| 493 | $UMULL r7,r6,r6 |
| 494 | $UMULH r8,r6,r6 |
| 495 | |
| 496 | addc r10,r7,r10 |
| 497 | adde r11,r8,r11 |
| 498 | addze r9,r0 |
| 499 | #sqr_add_c2(a,3,1,c2,c3,c1); |
| 500 | $LD r6,`3*$BNSZ`(r4) |
| 501 | $UMULL r7,r5,r6 |
| 502 | $UMULH r8,r5,r6 |
| 503 | |
| 504 | addc r10,r7,r10 |
| 505 | adde r11,r8,r11 |
| 506 | addze r9,r9 |
| 507 | |
| 508 | addc r10,r7,r10 |
| 509 | adde r11,r8,r11 |
| 510 | addze r9,r9 |
| 511 | #sqr_add_c2(a,4,0,c2,c3,c1); |
| 512 | $LD r5,`0*$BNSZ`(r4) |
| 513 | $LD r6,`4*$BNSZ`(r4) |
| 514 | $UMULL r7,r5,r6 |
| 515 | $UMULH r8,r5,r6 |
| 516 | |
| 517 | addc r10,r7,r10 |
| 518 | adde r11,r8,r11 |
| 519 | addze r9,r9 |
| 520 | |
| 521 | addc r10,r7,r10 |
| 522 | adde r11,r8,r11 |
| 523 | addze r9,r9 |
| 524 | $ST r10,`4*$BNSZ`(r3) #r[4]=c2; |
| 525 | #sqr_add_c2(a,5,0,c3,c1,c2); |
| 526 | $LD r6,`5*$BNSZ`(r4) |
| 527 | $UMULL r7,r5,r6 |
| 528 | $UMULH r8,r5,r6 |
| 529 | |
| 530 | addc r11,r7,r11 |
| 531 | adde r9,r8,r9 |
| 532 | addze r10,r0 |
| 533 | |
| 534 | addc r11,r7,r11 |
| 535 | adde r9,r8,r9 |
| 536 | addze r10,r10 |
| 537 | #sqr_add_c2(a,4,1,c3,c1,c2); |
| 538 | $LD r5,`1*$BNSZ`(r4) |
| 539 | $LD r6,`4*$BNSZ`(r4) |
| 540 | $UMULL r7,r5,r6 |
| 541 | $UMULH r8,r5,r6 |
| 542 | |
| 543 | addc r11,r7,r11 |
| 544 | adde r9,r8,r9 |
| 545 | addze r10,r10 |
| 546 | |
| 547 | addc r11,r7,r11 |
| 548 | adde r9,r8,r9 |
| 549 | addze r10,r10 |
| 550 | #sqr_add_c2(a,3,2,c3,c1,c2); |
| 551 | $LD r5,`2*$BNSZ`(r4) |
| 552 | $LD r6,`3*$BNSZ`(r4) |
| 553 | $UMULL r7,r5,r6 |
| 554 | $UMULH r8,r5,r6 |
| 555 | |
| 556 | addc r11,r7,r11 |
| 557 | adde r9,r8,r9 |
| 558 | addze r10,r10 |
| 559 | |
| 560 | addc r11,r7,r11 |
| 561 | adde r9,r8,r9 |
| 562 | addze r10,r10 |
| 563 | $ST r11,`5*$BNSZ`(r3) #r[5]=c3; |
| 564 | #sqr_add_c(a,3,c1,c2,c3); |
| 565 | $UMULL r7,r6,r6 |
| 566 | $UMULH r8,r6,r6 |
| 567 | addc r9,r7,r9 |
| 568 | adde r10,r8,r10 |
| 569 | addze r11,r0 |
| 570 | #sqr_add_c2(a,4,2,c1,c2,c3); |
| 571 | $LD r6,`4*$BNSZ`(r4) |
| 572 | $UMULL r7,r5,r6 |
| 573 | $UMULH r8,r5,r6 |
| 574 | |
| 575 | addc r9,r7,r9 |
| 576 | adde r10,r8,r10 |
| 577 | addze r11,r11 |
| 578 | |
| 579 | addc r9,r7,r9 |
| 580 | adde r10,r8,r10 |
| 581 | addze r11,r11 |
| 582 | #sqr_add_c2(a,5,1,c1,c2,c3); |
| 583 | $LD r5,`1*$BNSZ`(r4) |
| 584 | $LD r6,`5*$BNSZ`(r4) |
| 585 | $UMULL r7,r5,r6 |
| 586 | $UMULH r8,r5,r6 |
| 587 | |
| 588 | addc r9,r7,r9 |
| 589 | adde r10,r8,r10 |
| 590 | addze r11,r11 |
| 591 | |
| 592 | addc r9,r7,r9 |
| 593 | adde r10,r8,r10 |
| 594 | addze r11,r11 |
| 595 | #sqr_add_c2(a,6,0,c1,c2,c3); |
| 596 | $LD r5,`0*$BNSZ`(r4) |
| 597 | $LD r6,`6*$BNSZ`(r4) |
| 598 | $UMULL r7,r5,r6 |
| 599 | $UMULH r8,r5,r6 |
| 600 | addc r9,r7,r9 |
| 601 | adde r10,r8,r10 |
| 602 | addze r11,r11 |
| 603 | addc r9,r7,r9 |
| 604 | adde r10,r8,r10 |
| 605 | addze r11,r11 |
| 606 | $ST r9,`6*$BNSZ`(r3) #r[6]=c1; |
| 607 | #sqr_add_c2(a,7,0,c2,c3,c1); |
| 608 | $LD r6,`7*$BNSZ`(r4) |
| 609 | $UMULL r7,r5,r6 |
| 610 | $UMULH r8,r5,r6 |
| 611 | |
| 612 | addc r10,r7,r10 |
| 613 | adde r11,r8,r11 |
| 614 | addze r9,r0 |
| 615 | addc r10,r7,r10 |
| 616 | adde r11,r8,r11 |
| 617 | addze r9,r9 |
| 618 | #sqr_add_c2(a,6,1,c2,c3,c1); |
| 619 | $LD r5,`1*$BNSZ`(r4) |
| 620 | $LD r6,`6*$BNSZ`(r4) |
| 621 | $UMULL r7,r5,r6 |
| 622 | $UMULH r8,r5,r6 |
| 623 | |
| 624 | addc r10,r7,r10 |
| 625 | adde r11,r8,r11 |
| 626 | addze r9,r9 |
| 627 | addc r10,r7,r10 |
| 628 | adde r11,r8,r11 |
| 629 | addze r9,r9 |
| 630 | #sqr_add_c2(a,5,2,c2,c3,c1); |
| 631 | $LD r5,`2*$BNSZ`(r4) |
| 632 | $LD r6,`5*$BNSZ`(r4) |
| 633 | $UMULL r7,r5,r6 |
| 634 | $UMULH r8,r5,r6 |
| 635 | addc r10,r7,r10 |
| 636 | adde r11,r8,r11 |
| 637 | addze r9,r9 |
| 638 | addc r10,r7,r10 |
| 639 | adde r11,r8,r11 |
| 640 | addze r9,r9 |
| 641 | #sqr_add_c2(a,4,3,c2,c3,c1); |
| 642 | $LD r5,`3*$BNSZ`(r4) |
| 643 | $LD r6,`4*$BNSZ`(r4) |
| 644 | $UMULL r7,r5,r6 |
| 645 | $UMULH r8,r5,r6 |
| 646 | |
| 647 | addc r10,r7,r10 |
| 648 | adde r11,r8,r11 |
| 649 | addze r9,r9 |
| 650 | addc r10,r7,r10 |
| 651 | adde r11,r8,r11 |
| 652 | addze r9,r9 |
| 653 | $ST r10,`7*$BNSZ`(r3) #r[7]=c2; |
| 654 | #sqr_add_c(a,4,c3,c1,c2); |
| 655 | $UMULL r7,r6,r6 |
| 656 | $UMULH r8,r6,r6 |
| 657 | addc r11,r7,r11 |
| 658 | adde r9,r8,r9 |
| 659 | addze r10,r0 |
| 660 | #sqr_add_c2(a,5,3,c3,c1,c2); |
| 661 | $LD r6,`5*$BNSZ`(r4) |
| 662 | $UMULL r7,r5,r6 |
| 663 | $UMULH r8,r5,r6 |
| 664 | addc r11,r7,r11 |
| 665 | adde r9,r8,r9 |
| 666 | addze r10,r10 |
| 667 | addc r11,r7,r11 |
| 668 | adde r9,r8,r9 |
| 669 | addze r10,r10 |
| 670 | #sqr_add_c2(a,6,2,c3,c1,c2); |
| 671 | $LD r5,`2*$BNSZ`(r4) |
| 672 | $LD r6,`6*$BNSZ`(r4) |
| 673 | $UMULL r7,r5,r6 |
| 674 | $UMULH r8,r5,r6 |
| 675 | addc r11,r7,r11 |
| 676 | adde r9,r8,r9 |
| 677 | addze r10,r10 |
| 678 | |
| 679 | addc r11,r7,r11 |
| 680 | adde r9,r8,r9 |
| 681 | addze r10,r10 |
| 682 | #sqr_add_c2(a,7,1,c3,c1,c2); |
| 683 | $LD r5,`1*$BNSZ`(r4) |
| 684 | $LD r6,`7*$BNSZ`(r4) |
| 685 | $UMULL r7,r5,r6 |
| 686 | $UMULH r8,r5,r6 |
| 687 | addc r11,r7,r11 |
| 688 | adde r9,r8,r9 |
| 689 | addze r10,r10 |
| 690 | addc r11,r7,r11 |
| 691 | adde r9,r8,r9 |
| 692 | addze r10,r10 |
| 693 | $ST r11,`8*$BNSZ`(r3) #r[8]=c3; |
| 694 | #sqr_add_c2(a,7,2,c1,c2,c3); |
| 695 | $LD r5,`2*$BNSZ`(r4) |
| 696 | $UMULL r7,r5,r6 |
| 697 | $UMULH r8,r5,r6 |
| 698 | |
| 699 | addc r9,r7,r9 |
| 700 | adde r10,r8,r10 |
| 701 | addze r11,r0 |
| 702 | addc r9,r7,r9 |
| 703 | adde r10,r8,r10 |
| 704 | addze r11,r11 |
| 705 | #sqr_add_c2(a,6,3,c1,c2,c3); |
| 706 | $LD r5,`3*$BNSZ`(r4) |
| 707 | $LD r6,`6*$BNSZ`(r4) |
| 708 | $UMULL r7,r5,r6 |
| 709 | $UMULH r8,r5,r6 |
| 710 | addc r9,r7,r9 |
| 711 | adde r10,r8,r10 |
| 712 | addze r11,r11 |
| 713 | addc r9,r7,r9 |
| 714 | adde r10,r8,r10 |
| 715 | addze r11,r11 |
| 716 | #sqr_add_c2(a,5,4,c1,c2,c3); |
| 717 | $LD r5,`4*$BNSZ`(r4) |
| 718 | $LD r6,`5*$BNSZ`(r4) |
| 719 | $UMULL r7,r5,r6 |
| 720 | $UMULH r8,r5,r6 |
| 721 | addc r9,r7,r9 |
| 722 | adde r10,r8,r10 |
| 723 | addze r11,r11 |
| 724 | addc r9,r7,r9 |
| 725 | adde r10,r8,r10 |
| 726 | addze r11,r11 |
| 727 | $ST r9,`9*$BNSZ`(r3) #r[9]=c1; |
| 728 | #sqr_add_c(a,5,c2,c3,c1); |
| 729 | $UMULL r7,r6,r6 |
| 730 | $UMULH r8,r6,r6 |
| 731 | addc r10,r7,r10 |
| 732 | adde r11,r8,r11 |
| 733 | addze r9,r0 |
| 734 | #sqr_add_c2(a,6,4,c2,c3,c1); |
| 735 | $LD r6,`6*$BNSZ`(r4) |
| 736 | $UMULL r7,r5,r6 |
| 737 | $UMULH r8,r5,r6 |
| 738 | addc r10,r7,r10 |
| 739 | adde r11,r8,r11 |
| 740 | addze r9,r9 |
| 741 | addc r10,r7,r10 |
| 742 | adde r11,r8,r11 |
| 743 | addze r9,r9 |
| 744 | #sqr_add_c2(a,7,3,c2,c3,c1); |
| 745 | $LD r5,`3*$BNSZ`(r4) |
| 746 | $LD r6,`7*$BNSZ`(r4) |
| 747 | $UMULL r7,r5,r6 |
| 748 | $UMULH r8,r5,r6 |
| 749 | addc r10,r7,r10 |
| 750 | adde r11,r8,r11 |
| 751 | addze r9,r9 |
| 752 | addc r10,r7,r10 |
| 753 | adde r11,r8,r11 |
| 754 | addze r9,r9 |
| 755 | $ST r10,`10*$BNSZ`(r3) #r[10]=c2; |
| 756 | #sqr_add_c2(a,7,4,c3,c1,c2); |
| 757 | $LD r5,`4*$BNSZ`(r4) |
| 758 | $UMULL r7,r5,r6 |
| 759 | $UMULH r8,r5,r6 |
| 760 | addc r11,r7,r11 |
| 761 | adde r9,r8,r9 |
| 762 | addze r10,r0 |
| 763 | addc r11,r7,r11 |
| 764 | adde r9,r8,r9 |
| 765 | addze r10,r10 |
| 766 | #sqr_add_c2(a,6,5,c3,c1,c2); |
| 767 | $LD r5,`5*$BNSZ`(r4) |
| 768 | $LD r6,`6*$BNSZ`(r4) |
| 769 | $UMULL r7,r5,r6 |
| 770 | $UMULH r8,r5,r6 |
| 771 | addc r11,r7,r11 |
| 772 | adde r9,r8,r9 |
| 773 | addze r10,r10 |
| 774 | addc r11,r7,r11 |
| 775 | adde r9,r8,r9 |
| 776 | addze r10,r10 |
| 777 | $ST r11,`11*$BNSZ`(r3) #r[11]=c3; |
| 778 | #sqr_add_c(a,6,c1,c2,c3); |
| 779 | $UMULL r7,r6,r6 |
| 780 | $UMULH r8,r6,r6 |
| 781 | addc r9,r7,r9 |
| 782 | adde r10,r8,r10 |
| 783 | addze r11,r0 |
| 784 | #sqr_add_c2(a,7,5,c1,c2,c3) |
| 785 | $LD r6,`7*$BNSZ`(r4) |
| 786 | $UMULL r7,r5,r6 |
| 787 | $UMULH r8,r5,r6 |
| 788 | addc r9,r7,r9 |
| 789 | adde r10,r8,r10 |
| 790 | addze r11,r11 |
| 791 | addc r9,r7,r9 |
| 792 | adde r10,r8,r10 |
| 793 | addze r11,r11 |
| 794 | $ST r9,`12*$BNSZ`(r3) #r[12]=c1; |
| 795 | |
| 796 | #sqr_add_c2(a,7,6,c2,c3,c1) |
| 797 | $LD r5,`6*$BNSZ`(r4) |
| 798 | $UMULL r7,r5,r6 |
| 799 | $UMULH r8,r5,r6 |
| 800 | addc r10,r7,r10 |
| 801 | adde r11,r8,r11 |
| 802 | addze r9,r0 |
| 803 | addc r10,r7,r10 |
| 804 | adde r11,r8,r11 |
| 805 | addze r9,r9 |
| 806 | $ST r10,`13*$BNSZ`(r3) #r[13]=c2; |
| 807 | #sqr_add_c(a,7,c3,c1,c2); |
| 808 | $UMULL r7,r6,r6 |
| 809 | $UMULH r8,r6,r6 |
| 810 | addc r11,r7,r11 |
| 811 | adde r9,r8,r9 |
| 812 | $ST r11,`14*$BNSZ`(r3) #r[14]=c3; |
| 813 | $ST r9, `15*$BNSZ`(r3) #r[15]=c1; |
| 814 | |
| 815 | |
| 816 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 817 | |
| 818 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 819 | |
| 820 | # |
| 821 | # NOTE: The following label name should be changed to |
| 822 | # "bn_mul_comba4" i.e. remove the first dot |
| 823 | # for the gcc compiler. This should be automatically |
| 824 | # done in the build |
| 825 | # |
| 826 | |
| 827 | .align 4 |
| 828 | .bn_mul_comba4: |
| 829 | # |
| 830 | # This is an optimized version of the bn_mul_comba4 routine. |
| 831 | # |
| 832 | # void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b) |
| 833 | # r3 contains r |
| 834 | # r4 contains a |
| 835 | # r5 contains b |
| 836 | # r6, r7 are the 2 BN_ULONGs being multiplied. |
| 837 | # r8, r9 are the results of the 32x32 giving 64 multiply. |
| 838 | # r10, r11, r12 are the equivalents of c1, c2, and c3. |
| 839 | # |
| 840 | xor r0,r0,r0 #r0=0. Used in addze below. |
| 841 | #mul_add_c(a[0],b[0],c1,c2,c3); |
| 842 | $LD r6,`0*$BNSZ`(r4) |
| 843 | $LD r7,`0*$BNSZ`(r5) |
| 844 | $UMULL r10,r6,r7 |
| 845 | $UMULH r11,r6,r7 |
| 846 | $ST r10,`0*$BNSZ`(r3) #r[0]=c1 |
| 847 | #mul_add_c(a[0],b[1],c2,c3,c1); |
| 848 | $LD r7,`1*$BNSZ`(r5) |
| 849 | $UMULL r8,r6,r7 |
| 850 | $UMULH r9,r6,r7 |
| 851 | addc r11,r8,r11 |
| 852 | adde r12,r9,r0 |
| 853 | addze r10,r0 |
| 854 | #mul_add_c(a[1],b[0],c2,c3,c1); |
| 855 | $LD r6, `1*$BNSZ`(r4) |
| 856 | $LD r7, `0*$BNSZ`(r5) |
| 857 | $UMULL r8,r6,r7 |
| 858 | $UMULH r9,r6,r7 |
| 859 | addc r11,r8,r11 |
| 860 | adde r12,r9,r12 |
| 861 | addze r10,r10 |
| 862 | $ST r11,`1*$BNSZ`(r3) #r[1]=c2 |
| 863 | #mul_add_c(a[2],b[0],c3,c1,c2); |
| 864 | $LD r6,`2*$BNSZ`(r4) |
| 865 | $UMULL r8,r6,r7 |
| 866 | $UMULH r9,r6,r7 |
| 867 | addc r12,r8,r12 |
| 868 | adde r10,r9,r10 |
| 869 | addze r11,r0 |
| 870 | #mul_add_c(a[1],b[1],c3,c1,c2); |
| 871 | $LD r6,`1*$BNSZ`(r4) |
| 872 | $LD r7,`1*$BNSZ`(r5) |
| 873 | $UMULL r8,r6,r7 |
| 874 | $UMULH r9,r6,r7 |
| 875 | addc r12,r8,r12 |
| 876 | adde r10,r9,r10 |
| 877 | addze r11,r11 |
| 878 | #mul_add_c(a[0],b[2],c3,c1,c2); |
| 879 | $LD r6,`0*$BNSZ`(r4) |
| 880 | $LD r7,`2*$BNSZ`(r5) |
| 881 | $UMULL r8,r6,r7 |
| 882 | $UMULH r9,r6,r7 |
| 883 | addc r12,r8,r12 |
| 884 | adde r10,r9,r10 |
| 885 | addze r11,r11 |
| 886 | $ST r12,`2*$BNSZ`(r3) #r[2]=c3 |
| 887 | #mul_add_c(a[0],b[3],c1,c2,c3); |
| 888 | $LD r7,`3*$BNSZ`(r5) |
| 889 | $UMULL r8,r6,r7 |
| 890 | $UMULH r9,r6,r7 |
| 891 | addc r10,r8,r10 |
| 892 | adde r11,r9,r11 |
| 893 | addze r12,r0 |
| 894 | #mul_add_c(a[1],b[2],c1,c2,c3); |
| 895 | $LD r6,`1*$BNSZ`(r4) |
| 896 | $LD r7,`2*$BNSZ`(r5) |
| 897 | $UMULL r8,r6,r7 |
| 898 | $UMULH r9,r6,r7 |
| 899 | addc r10,r8,r10 |
| 900 | adde r11,r9,r11 |
| 901 | addze r12,r12 |
| 902 | #mul_add_c(a[2],b[1],c1,c2,c3); |
| 903 | $LD r6,`2*$BNSZ`(r4) |
| 904 | $LD r7,`1*$BNSZ`(r5) |
| 905 | $UMULL r8,r6,r7 |
| 906 | $UMULH r9,r6,r7 |
| 907 | addc r10,r8,r10 |
| 908 | adde r11,r9,r11 |
| 909 | addze r12,r12 |
| 910 | #mul_add_c(a[3],b[0],c1,c2,c3); |
| 911 | $LD r6,`3*$BNSZ`(r4) |
| 912 | $LD r7,`0*$BNSZ`(r5) |
| 913 | $UMULL r8,r6,r7 |
| 914 | $UMULH r9,r6,r7 |
| 915 | addc r10,r8,r10 |
| 916 | adde r11,r9,r11 |
| 917 | addze r12,r12 |
| 918 | $ST r10,`3*$BNSZ`(r3) #r[3]=c1 |
| 919 | #mul_add_c(a[3],b[1],c2,c3,c1); |
| 920 | $LD r7,`1*$BNSZ`(r5) |
| 921 | $UMULL r8,r6,r7 |
| 922 | $UMULH r9,r6,r7 |
| 923 | addc r11,r8,r11 |
| 924 | adde r12,r9,r12 |
| 925 | addze r10,r0 |
| 926 | #mul_add_c(a[2],b[2],c2,c3,c1); |
| 927 | $LD r6,`2*$BNSZ`(r4) |
| 928 | $LD r7,`2*$BNSZ`(r5) |
| 929 | $UMULL r8,r6,r7 |
| 930 | $UMULH r9,r6,r7 |
| 931 | addc r11,r8,r11 |
| 932 | adde r12,r9,r12 |
| 933 | addze r10,r10 |
| 934 | #mul_add_c(a[1],b[3],c2,c3,c1); |
| 935 | $LD r6,`1*$BNSZ`(r4) |
| 936 | $LD r7,`3*$BNSZ`(r5) |
| 937 | $UMULL r8,r6,r7 |
| 938 | $UMULH r9,r6,r7 |
| 939 | addc r11,r8,r11 |
| 940 | adde r12,r9,r12 |
| 941 | addze r10,r10 |
| 942 | $ST r11,`4*$BNSZ`(r3) #r[4]=c2 |
| 943 | #mul_add_c(a[2],b[3],c3,c1,c2); |
| 944 | $LD r6,`2*$BNSZ`(r4) |
| 945 | $UMULL r8,r6,r7 |
| 946 | $UMULH r9,r6,r7 |
| 947 | addc r12,r8,r12 |
| 948 | adde r10,r9,r10 |
| 949 | addze r11,r0 |
| 950 | #mul_add_c(a[3],b[2],c3,c1,c2); |
| 951 | $LD r6,`3*$BNSZ`(r4) |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 952 | $LD r7,`2*$BNSZ`(r4) |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 953 | $UMULL r8,r6,r7 |
| 954 | $UMULH r9,r6,r7 |
| 955 | addc r12,r8,r12 |
| 956 | adde r10,r9,r10 |
| 957 | addze r11,r11 |
| 958 | $ST r12,`5*$BNSZ`(r3) #r[5]=c3 |
| 959 | #mul_add_c(a[3],b[3],c1,c2,c3); |
| 960 | $LD r7,`3*$BNSZ`(r5) |
| 961 | $UMULL r8,r6,r7 |
| 962 | $UMULH r9,r6,r7 |
| 963 | addc r10,r8,r10 |
| 964 | adde r11,r9,r11 |
| 965 | |
| 966 | $ST r10,`6*$BNSZ`(r3) #r[6]=c1 |
| 967 | $ST r11,`7*$BNSZ`(r3) #r[7]=c2 |
| 968 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 969 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 970 | |
| 971 | # |
| 972 | # NOTE: The following label name should be changed to |
| 973 | # "bn_mul_comba8" i.e. remove the first dot |
| 974 | # for the gcc compiler. This should be automatically |
| 975 | # done in the build |
| 976 | # |
| 977 | |
| 978 | .align 4 |
| 979 | .bn_mul_comba8: |
| 980 | # |
| 981 | # Optimized version of the bn_mul_comba8 routine. |
| 982 | # |
| 983 | # void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b) |
| 984 | # r3 contains r |
| 985 | # r4 contains a |
| 986 | # r5 contains b |
| 987 | # r6, r7 are the 2 BN_ULONGs being multiplied. |
| 988 | # r8, r9 are the results of the 32x32 giving 64 multiply. |
| 989 | # r10, r11, r12 are the equivalents of c1, c2, and c3. |
| 990 | # |
| 991 | xor r0,r0,r0 #r0=0. Used in addze below. |
| 992 | |
| 993 | #mul_add_c(a[0],b[0],c1,c2,c3); |
| 994 | $LD r6,`0*$BNSZ`(r4) #a[0] |
| 995 | $LD r7,`0*$BNSZ`(r5) #b[0] |
| 996 | $UMULL r10,r6,r7 |
| 997 | $UMULH r11,r6,r7 |
| 998 | $ST r10,`0*$BNSZ`(r3) #r[0]=c1; |
| 999 | #mul_add_c(a[0],b[1],c2,c3,c1); |
| 1000 | $LD r7,`1*$BNSZ`(r5) |
| 1001 | $UMULL r8,r6,r7 |
| 1002 | $UMULH r9,r6,r7 |
| 1003 | addc r11,r11,r8 |
| 1004 | addze r12,r9 # since we didnt set r12 to zero before. |
| 1005 | addze r10,r0 |
| 1006 | #mul_add_c(a[1],b[0],c2,c3,c1); |
| 1007 | $LD r6,`1*$BNSZ`(r4) |
| 1008 | $LD r7,`0*$BNSZ`(r5) |
| 1009 | $UMULL r8,r6,r7 |
| 1010 | $UMULH r9,r6,r7 |
| 1011 | addc r11,r11,r8 |
| 1012 | adde r12,r12,r9 |
| 1013 | addze r10,r10 |
| 1014 | $ST r11,`1*$BNSZ`(r3) #r[1]=c2; |
| 1015 | #mul_add_c(a[2],b[0],c3,c1,c2); |
| 1016 | $LD r6,`2*$BNSZ`(r4) |
| 1017 | $UMULL r8,r6,r7 |
| 1018 | $UMULH r9,r6,r7 |
| 1019 | addc r12,r12,r8 |
| 1020 | adde r10,r10,r9 |
| 1021 | addze r11,r0 |
| 1022 | #mul_add_c(a[1],b[1],c3,c1,c2); |
| 1023 | $LD r6,`1*$BNSZ`(r4) |
| 1024 | $LD r7,`1*$BNSZ`(r5) |
| 1025 | $UMULL r8,r6,r7 |
| 1026 | $UMULH r9,r6,r7 |
| 1027 | addc r12,r12,r8 |
| 1028 | adde r10,r10,r9 |
| 1029 | addze r11,r11 |
| 1030 | #mul_add_c(a[0],b[2],c3,c1,c2); |
| 1031 | $LD r6,`0*$BNSZ`(r4) |
| 1032 | $LD r7,`2*$BNSZ`(r5) |
| 1033 | $UMULL r8,r6,r7 |
| 1034 | $UMULH r9,r6,r7 |
| 1035 | addc r12,r12,r8 |
| 1036 | adde r10,r10,r9 |
| 1037 | addze r11,r11 |
| 1038 | $ST r12,`2*$BNSZ`(r3) #r[2]=c3; |
| 1039 | #mul_add_c(a[0],b[3],c1,c2,c3); |
| 1040 | $LD r7,`3*$BNSZ`(r5) |
| 1041 | $UMULL r8,r6,r7 |
| 1042 | $UMULH r9,r6,r7 |
| 1043 | addc r10,r10,r8 |
| 1044 | adde r11,r11,r9 |
| 1045 | addze r12,r0 |
| 1046 | #mul_add_c(a[1],b[2],c1,c2,c3); |
| 1047 | $LD r6,`1*$BNSZ`(r4) |
| 1048 | $LD r7,`2*$BNSZ`(r5) |
| 1049 | $UMULL r8,r6,r7 |
| 1050 | $UMULH r9,r6,r7 |
| 1051 | addc r10,r10,r8 |
| 1052 | adde r11,r11,r9 |
| 1053 | addze r12,r12 |
| 1054 | |
| 1055 | #mul_add_c(a[2],b[1],c1,c2,c3); |
| 1056 | $LD r6,`2*$BNSZ`(r4) |
| 1057 | $LD r7,`1*$BNSZ`(r5) |
| 1058 | $UMULL r8,r6,r7 |
| 1059 | $UMULH r9,r6,r7 |
| 1060 | addc r10,r10,r8 |
| 1061 | adde r11,r11,r9 |
| 1062 | addze r12,r12 |
| 1063 | #mul_add_c(a[3],b[0],c1,c2,c3); |
| 1064 | $LD r6,`3*$BNSZ`(r4) |
| 1065 | $LD r7,`0*$BNSZ`(r5) |
| 1066 | $UMULL r8,r6,r7 |
| 1067 | $UMULH r9,r6,r7 |
| 1068 | addc r10,r10,r8 |
| 1069 | adde r11,r11,r9 |
| 1070 | addze r12,r12 |
| 1071 | $ST r10,`3*$BNSZ`(r3) #r[3]=c1; |
| 1072 | #mul_add_c(a[4],b[0],c2,c3,c1); |
| 1073 | $LD r6,`4*$BNSZ`(r4) |
| 1074 | $UMULL r8,r6,r7 |
| 1075 | $UMULH r9,r6,r7 |
| 1076 | addc r11,r11,r8 |
| 1077 | adde r12,r12,r9 |
| 1078 | addze r10,r0 |
| 1079 | #mul_add_c(a[3],b[1],c2,c3,c1); |
| 1080 | $LD r6,`3*$BNSZ`(r4) |
| 1081 | $LD r7,`1*$BNSZ`(r5) |
| 1082 | $UMULL r8,r6,r7 |
| 1083 | $UMULH r9,r6,r7 |
| 1084 | addc r11,r11,r8 |
| 1085 | adde r12,r12,r9 |
| 1086 | addze r10,r10 |
| 1087 | #mul_add_c(a[2],b[2],c2,c3,c1); |
| 1088 | $LD r6,`2*$BNSZ`(r4) |
| 1089 | $LD r7,`2*$BNSZ`(r5) |
| 1090 | $UMULL r8,r6,r7 |
| 1091 | $UMULH r9,r6,r7 |
| 1092 | addc r11,r11,r8 |
| 1093 | adde r12,r12,r9 |
| 1094 | addze r10,r10 |
| 1095 | #mul_add_c(a[1],b[3],c2,c3,c1); |
| 1096 | $LD r6,`1*$BNSZ`(r4) |
| 1097 | $LD r7,`3*$BNSZ`(r5) |
| 1098 | $UMULL r8,r6,r7 |
| 1099 | $UMULH r9,r6,r7 |
| 1100 | addc r11,r11,r8 |
| 1101 | adde r12,r12,r9 |
| 1102 | addze r10,r10 |
| 1103 | #mul_add_c(a[0],b[4],c2,c3,c1); |
| 1104 | $LD r6,`0*$BNSZ`(r4) |
| 1105 | $LD r7,`4*$BNSZ`(r5) |
| 1106 | $UMULL r8,r6,r7 |
| 1107 | $UMULH r9,r6,r7 |
| 1108 | addc r11,r11,r8 |
| 1109 | adde r12,r12,r9 |
| 1110 | addze r10,r10 |
| 1111 | $ST r11,`4*$BNSZ`(r3) #r[4]=c2; |
| 1112 | #mul_add_c(a[0],b[5],c3,c1,c2); |
| 1113 | $LD r7,`5*$BNSZ`(r5) |
| 1114 | $UMULL r8,r6,r7 |
| 1115 | $UMULH r9,r6,r7 |
| 1116 | addc r12,r12,r8 |
| 1117 | adde r10,r10,r9 |
| 1118 | addze r11,r0 |
| 1119 | #mul_add_c(a[1],b[4],c3,c1,c2); |
| 1120 | $LD r6,`1*$BNSZ`(r4) |
| 1121 | $LD r7,`4*$BNSZ`(r5) |
| 1122 | $UMULL r8,r6,r7 |
| 1123 | $UMULH r9,r6,r7 |
| 1124 | addc r12,r12,r8 |
| 1125 | adde r10,r10,r9 |
| 1126 | addze r11,r11 |
| 1127 | #mul_add_c(a[2],b[3],c3,c1,c2); |
| 1128 | $LD r6,`2*$BNSZ`(r4) |
| 1129 | $LD r7,`3*$BNSZ`(r5) |
| 1130 | $UMULL r8,r6,r7 |
| 1131 | $UMULH r9,r6,r7 |
| 1132 | addc r12,r12,r8 |
| 1133 | adde r10,r10,r9 |
| 1134 | addze r11,r11 |
| 1135 | #mul_add_c(a[3],b[2],c3,c1,c2); |
| 1136 | $LD r6,`3*$BNSZ`(r4) |
| 1137 | $LD r7,`2*$BNSZ`(r5) |
| 1138 | $UMULL r8,r6,r7 |
| 1139 | $UMULH r9,r6,r7 |
| 1140 | addc r12,r12,r8 |
| 1141 | adde r10,r10,r9 |
| 1142 | addze r11,r11 |
| 1143 | #mul_add_c(a[4],b[1],c3,c1,c2); |
| 1144 | $LD r6,`4*$BNSZ`(r4) |
| 1145 | $LD r7,`1*$BNSZ`(r5) |
| 1146 | $UMULL r8,r6,r7 |
| 1147 | $UMULH r9,r6,r7 |
| 1148 | addc r12,r12,r8 |
| 1149 | adde r10,r10,r9 |
| 1150 | addze r11,r11 |
| 1151 | #mul_add_c(a[5],b[0],c3,c1,c2); |
| 1152 | $LD r6,`5*$BNSZ`(r4) |
| 1153 | $LD r7,`0*$BNSZ`(r5) |
| 1154 | $UMULL r8,r6,r7 |
| 1155 | $UMULH r9,r6,r7 |
| 1156 | addc r12,r12,r8 |
| 1157 | adde r10,r10,r9 |
| 1158 | addze r11,r11 |
| 1159 | $ST r12,`5*$BNSZ`(r3) #r[5]=c3; |
| 1160 | #mul_add_c(a[6],b[0],c1,c2,c3); |
| 1161 | $LD r6,`6*$BNSZ`(r4) |
| 1162 | $UMULL r8,r6,r7 |
| 1163 | $UMULH r9,r6,r7 |
| 1164 | addc r10,r10,r8 |
| 1165 | adde r11,r11,r9 |
| 1166 | addze r12,r0 |
| 1167 | #mul_add_c(a[5],b[1],c1,c2,c3); |
| 1168 | $LD r6,`5*$BNSZ`(r4) |
| 1169 | $LD r7,`1*$BNSZ`(r5) |
| 1170 | $UMULL r8,r6,r7 |
| 1171 | $UMULH r9,r6,r7 |
| 1172 | addc r10,r10,r8 |
| 1173 | adde r11,r11,r9 |
| 1174 | addze r12,r12 |
| 1175 | #mul_add_c(a[4],b[2],c1,c2,c3); |
| 1176 | $LD r6,`4*$BNSZ`(r4) |
| 1177 | $LD r7,`2*$BNSZ`(r5) |
| 1178 | $UMULL r8,r6,r7 |
| 1179 | $UMULH r9,r6,r7 |
| 1180 | addc r10,r10,r8 |
| 1181 | adde r11,r11,r9 |
| 1182 | addze r12,r12 |
| 1183 | #mul_add_c(a[3],b[3],c1,c2,c3); |
| 1184 | $LD r6,`3*$BNSZ`(r4) |
| 1185 | $LD r7,`3*$BNSZ`(r5) |
| 1186 | $UMULL r8,r6,r7 |
| 1187 | $UMULH r9,r6,r7 |
| 1188 | addc r10,r10,r8 |
| 1189 | adde r11,r11,r9 |
| 1190 | addze r12,r12 |
| 1191 | #mul_add_c(a[2],b[4],c1,c2,c3); |
| 1192 | $LD r6,`2*$BNSZ`(r4) |
| 1193 | $LD r7,`4*$BNSZ`(r5) |
| 1194 | $UMULL r8,r6,r7 |
| 1195 | $UMULH r9,r6,r7 |
| 1196 | addc r10,r10,r8 |
| 1197 | adde r11,r11,r9 |
| 1198 | addze r12,r12 |
| 1199 | #mul_add_c(a[1],b[5],c1,c2,c3); |
| 1200 | $LD r6,`1*$BNSZ`(r4) |
| 1201 | $LD r7,`5*$BNSZ`(r5) |
| 1202 | $UMULL r8,r6,r7 |
| 1203 | $UMULH r9,r6,r7 |
| 1204 | addc r10,r10,r8 |
| 1205 | adde r11,r11,r9 |
| 1206 | addze r12,r12 |
| 1207 | #mul_add_c(a[0],b[6],c1,c2,c3); |
| 1208 | $LD r6,`0*$BNSZ`(r4) |
| 1209 | $LD r7,`6*$BNSZ`(r5) |
| 1210 | $UMULL r8,r6,r7 |
| 1211 | $UMULH r9,r6,r7 |
| 1212 | addc r10,r10,r8 |
| 1213 | adde r11,r11,r9 |
| 1214 | addze r12,r12 |
| 1215 | $ST r10,`6*$BNSZ`(r3) #r[6]=c1; |
| 1216 | #mul_add_c(a[0],b[7],c2,c3,c1); |
| 1217 | $LD r7,`7*$BNSZ`(r5) |
| 1218 | $UMULL r8,r6,r7 |
| 1219 | $UMULH r9,r6,r7 |
| 1220 | addc r11,r11,r8 |
| 1221 | adde r12,r12,r9 |
| 1222 | addze r10,r0 |
| 1223 | #mul_add_c(a[1],b[6],c2,c3,c1); |
| 1224 | $LD r6,`1*$BNSZ`(r4) |
| 1225 | $LD r7,`6*$BNSZ`(r5) |
| 1226 | $UMULL r8,r6,r7 |
| 1227 | $UMULH r9,r6,r7 |
| 1228 | addc r11,r11,r8 |
| 1229 | adde r12,r12,r9 |
| 1230 | addze r10,r10 |
| 1231 | #mul_add_c(a[2],b[5],c2,c3,c1); |
| 1232 | $LD r6,`2*$BNSZ`(r4) |
| 1233 | $LD r7,`5*$BNSZ`(r5) |
| 1234 | $UMULL r8,r6,r7 |
| 1235 | $UMULH r9,r6,r7 |
| 1236 | addc r11,r11,r8 |
| 1237 | adde r12,r12,r9 |
| 1238 | addze r10,r10 |
| 1239 | #mul_add_c(a[3],b[4],c2,c3,c1); |
| 1240 | $LD r6,`3*$BNSZ`(r4) |
| 1241 | $LD r7,`4*$BNSZ`(r5) |
| 1242 | $UMULL r8,r6,r7 |
| 1243 | $UMULH r9,r6,r7 |
| 1244 | addc r11,r11,r8 |
| 1245 | adde r12,r12,r9 |
| 1246 | addze r10,r10 |
| 1247 | #mul_add_c(a[4],b[3],c2,c3,c1); |
| 1248 | $LD r6,`4*$BNSZ`(r4) |
| 1249 | $LD r7,`3*$BNSZ`(r5) |
| 1250 | $UMULL r8,r6,r7 |
| 1251 | $UMULH r9,r6,r7 |
| 1252 | addc r11,r11,r8 |
| 1253 | adde r12,r12,r9 |
| 1254 | addze r10,r10 |
| 1255 | #mul_add_c(a[5],b[2],c2,c3,c1); |
| 1256 | $LD r6,`5*$BNSZ`(r4) |
| 1257 | $LD r7,`2*$BNSZ`(r5) |
| 1258 | $UMULL r8,r6,r7 |
| 1259 | $UMULH r9,r6,r7 |
| 1260 | addc r11,r11,r8 |
| 1261 | adde r12,r12,r9 |
| 1262 | addze r10,r10 |
| 1263 | #mul_add_c(a[6],b[1],c2,c3,c1); |
| 1264 | $LD r6,`6*$BNSZ`(r4) |
| 1265 | $LD r7,`1*$BNSZ`(r5) |
| 1266 | $UMULL r8,r6,r7 |
| 1267 | $UMULH r9,r6,r7 |
| 1268 | addc r11,r11,r8 |
| 1269 | adde r12,r12,r9 |
| 1270 | addze r10,r10 |
| 1271 | #mul_add_c(a[7],b[0],c2,c3,c1); |
| 1272 | $LD r6,`7*$BNSZ`(r4) |
| 1273 | $LD r7,`0*$BNSZ`(r5) |
| 1274 | $UMULL r8,r6,r7 |
| 1275 | $UMULH r9,r6,r7 |
| 1276 | addc r11,r11,r8 |
| 1277 | adde r12,r12,r9 |
| 1278 | addze r10,r10 |
| 1279 | $ST r11,`7*$BNSZ`(r3) #r[7]=c2; |
| 1280 | #mul_add_c(a[7],b[1],c3,c1,c2); |
| 1281 | $LD r7,`1*$BNSZ`(r5) |
| 1282 | $UMULL r8,r6,r7 |
| 1283 | $UMULH r9,r6,r7 |
| 1284 | addc r12,r12,r8 |
| 1285 | adde r10,r10,r9 |
| 1286 | addze r11,r0 |
| 1287 | #mul_add_c(a[6],b[2],c3,c1,c2); |
| 1288 | $LD r6,`6*$BNSZ`(r4) |
| 1289 | $LD r7,`2*$BNSZ`(r5) |
| 1290 | $UMULL r8,r6,r7 |
| 1291 | $UMULH r9,r6,r7 |
| 1292 | addc r12,r12,r8 |
| 1293 | adde r10,r10,r9 |
| 1294 | addze r11,r11 |
| 1295 | #mul_add_c(a[5],b[3],c3,c1,c2); |
| 1296 | $LD r6,`5*$BNSZ`(r4) |
| 1297 | $LD r7,`3*$BNSZ`(r5) |
| 1298 | $UMULL r8,r6,r7 |
| 1299 | $UMULH r9,r6,r7 |
| 1300 | addc r12,r12,r8 |
| 1301 | adde r10,r10,r9 |
| 1302 | addze r11,r11 |
| 1303 | #mul_add_c(a[4],b[4],c3,c1,c2); |
| 1304 | $LD r6,`4*$BNSZ`(r4) |
| 1305 | $LD r7,`4*$BNSZ`(r5) |
| 1306 | $UMULL r8,r6,r7 |
| 1307 | $UMULH r9,r6,r7 |
| 1308 | addc r12,r12,r8 |
| 1309 | adde r10,r10,r9 |
| 1310 | addze r11,r11 |
| 1311 | #mul_add_c(a[3],b[5],c3,c1,c2); |
| 1312 | $LD r6,`3*$BNSZ`(r4) |
| 1313 | $LD r7,`5*$BNSZ`(r5) |
| 1314 | $UMULL r8,r6,r7 |
| 1315 | $UMULH r9,r6,r7 |
| 1316 | addc r12,r12,r8 |
| 1317 | adde r10,r10,r9 |
| 1318 | addze r11,r11 |
| 1319 | #mul_add_c(a[2],b[6],c3,c1,c2); |
| 1320 | $LD r6,`2*$BNSZ`(r4) |
| 1321 | $LD r7,`6*$BNSZ`(r5) |
| 1322 | $UMULL r8,r6,r7 |
| 1323 | $UMULH r9,r6,r7 |
| 1324 | addc r12,r12,r8 |
| 1325 | adde r10,r10,r9 |
| 1326 | addze r11,r11 |
| 1327 | #mul_add_c(a[1],b[7],c3,c1,c2); |
| 1328 | $LD r6,`1*$BNSZ`(r4) |
| 1329 | $LD r7,`7*$BNSZ`(r5) |
| 1330 | $UMULL r8,r6,r7 |
| 1331 | $UMULH r9,r6,r7 |
| 1332 | addc r12,r12,r8 |
| 1333 | adde r10,r10,r9 |
| 1334 | addze r11,r11 |
| 1335 | $ST r12,`8*$BNSZ`(r3) #r[8]=c3; |
| 1336 | #mul_add_c(a[2],b[7],c1,c2,c3); |
| 1337 | $LD r6,`2*$BNSZ`(r4) |
| 1338 | $UMULL r8,r6,r7 |
| 1339 | $UMULH r9,r6,r7 |
| 1340 | addc r10,r10,r8 |
| 1341 | adde r11,r11,r9 |
| 1342 | addze r12,r0 |
| 1343 | #mul_add_c(a[3],b[6],c1,c2,c3); |
| 1344 | $LD r6,`3*$BNSZ`(r4) |
| 1345 | $LD r7,`6*$BNSZ`(r5) |
| 1346 | $UMULL r8,r6,r7 |
| 1347 | $UMULH r9,r6,r7 |
| 1348 | addc r10,r10,r8 |
| 1349 | adde r11,r11,r9 |
| 1350 | addze r12,r12 |
| 1351 | #mul_add_c(a[4],b[5],c1,c2,c3); |
| 1352 | $LD r6,`4*$BNSZ`(r4) |
| 1353 | $LD r7,`5*$BNSZ`(r5) |
| 1354 | $UMULL r8,r6,r7 |
| 1355 | $UMULH r9,r6,r7 |
| 1356 | addc r10,r10,r8 |
| 1357 | adde r11,r11,r9 |
| 1358 | addze r12,r12 |
| 1359 | #mul_add_c(a[5],b[4],c1,c2,c3); |
| 1360 | $LD r6,`5*$BNSZ`(r4) |
| 1361 | $LD r7,`4*$BNSZ`(r5) |
| 1362 | $UMULL r8,r6,r7 |
| 1363 | $UMULH r9,r6,r7 |
| 1364 | addc r10,r10,r8 |
| 1365 | adde r11,r11,r9 |
| 1366 | addze r12,r12 |
| 1367 | #mul_add_c(a[6],b[3],c1,c2,c3); |
| 1368 | $LD r6,`6*$BNSZ`(r4) |
| 1369 | $LD r7,`3*$BNSZ`(r5) |
| 1370 | $UMULL r8,r6,r7 |
| 1371 | $UMULH r9,r6,r7 |
| 1372 | addc r10,r10,r8 |
| 1373 | adde r11,r11,r9 |
| 1374 | addze r12,r12 |
| 1375 | #mul_add_c(a[7],b[2],c1,c2,c3); |
| 1376 | $LD r6,`7*$BNSZ`(r4) |
| 1377 | $LD r7,`2*$BNSZ`(r5) |
| 1378 | $UMULL r8,r6,r7 |
| 1379 | $UMULH r9,r6,r7 |
| 1380 | addc r10,r10,r8 |
| 1381 | adde r11,r11,r9 |
| 1382 | addze r12,r12 |
| 1383 | $ST r10,`9*$BNSZ`(r3) #r[9]=c1; |
| 1384 | #mul_add_c(a[7],b[3],c2,c3,c1); |
| 1385 | $LD r7,`3*$BNSZ`(r5) |
| 1386 | $UMULL r8,r6,r7 |
| 1387 | $UMULH r9,r6,r7 |
| 1388 | addc r11,r11,r8 |
| 1389 | adde r12,r12,r9 |
| 1390 | addze r10,r0 |
| 1391 | #mul_add_c(a[6],b[4],c2,c3,c1); |
| 1392 | $LD r6,`6*$BNSZ`(r4) |
| 1393 | $LD r7,`4*$BNSZ`(r5) |
| 1394 | $UMULL r8,r6,r7 |
| 1395 | $UMULH r9,r6,r7 |
| 1396 | addc r11,r11,r8 |
| 1397 | adde r12,r12,r9 |
| 1398 | addze r10,r10 |
| 1399 | #mul_add_c(a[5],b[5],c2,c3,c1); |
| 1400 | $LD r6,`5*$BNSZ`(r4) |
| 1401 | $LD r7,`5*$BNSZ`(r5) |
| 1402 | $UMULL r8,r6,r7 |
| 1403 | $UMULH r9,r6,r7 |
| 1404 | addc r11,r11,r8 |
| 1405 | adde r12,r12,r9 |
| 1406 | addze r10,r10 |
| 1407 | #mul_add_c(a[4],b[6],c2,c3,c1); |
| 1408 | $LD r6,`4*$BNSZ`(r4) |
| 1409 | $LD r7,`6*$BNSZ`(r5) |
| 1410 | $UMULL r8,r6,r7 |
| 1411 | $UMULH r9,r6,r7 |
| 1412 | addc r11,r11,r8 |
| 1413 | adde r12,r12,r9 |
| 1414 | addze r10,r10 |
| 1415 | #mul_add_c(a[3],b[7],c2,c3,c1); |
| 1416 | $LD r6,`3*$BNSZ`(r4) |
| 1417 | $LD r7,`7*$BNSZ`(r5) |
| 1418 | $UMULL r8,r6,r7 |
| 1419 | $UMULH r9,r6,r7 |
| 1420 | addc r11,r11,r8 |
| 1421 | adde r12,r12,r9 |
| 1422 | addze r10,r10 |
| 1423 | $ST r11,`10*$BNSZ`(r3) #r[10]=c2; |
| 1424 | #mul_add_c(a[4],b[7],c3,c1,c2); |
| 1425 | $LD r6,`4*$BNSZ`(r4) |
| 1426 | $UMULL r8,r6,r7 |
| 1427 | $UMULH r9,r6,r7 |
| 1428 | addc r12,r12,r8 |
| 1429 | adde r10,r10,r9 |
| 1430 | addze r11,r0 |
| 1431 | #mul_add_c(a[5],b[6],c3,c1,c2); |
| 1432 | $LD r6,`5*$BNSZ`(r4) |
| 1433 | $LD r7,`6*$BNSZ`(r5) |
| 1434 | $UMULL r8,r6,r7 |
| 1435 | $UMULH r9,r6,r7 |
| 1436 | addc r12,r12,r8 |
| 1437 | adde r10,r10,r9 |
| 1438 | addze r11,r11 |
| 1439 | #mul_add_c(a[6],b[5],c3,c1,c2); |
| 1440 | $LD r6,`6*$BNSZ`(r4) |
| 1441 | $LD r7,`5*$BNSZ`(r5) |
| 1442 | $UMULL r8,r6,r7 |
| 1443 | $UMULH r9,r6,r7 |
| 1444 | addc r12,r12,r8 |
| 1445 | adde r10,r10,r9 |
| 1446 | addze r11,r11 |
| 1447 | #mul_add_c(a[7],b[4],c3,c1,c2); |
| 1448 | $LD r6,`7*$BNSZ`(r4) |
| 1449 | $LD r7,`4*$BNSZ`(r5) |
| 1450 | $UMULL r8,r6,r7 |
| 1451 | $UMULH r9,r6,r7 |
| 1452 | addc r12,r12,r8 |
| 1453 | adde r10,r10,r9 |
| 1454 | addze r11,r11 |
| 1455 | $ST r12,`11*$BNSZ`(r3) #r[11]=c3; |
| 1456 | #mul_add_c(a[7],b[5],c1,c2,c3); |
| 1457 | $LD r7,`5*$BNSZ`(r5) |
| 1458 | $UMULL r8,r6,r7 |
| 1459 | $UMULH r9,r6,r7 |
| 1460 | addc r10,r10,r8 |
| 1461 | adde r11,r11,r9 |
| 1462 | addze r12,r0 |
| 1463 | #mul_add_c(a[6],b[6],c1,c2,c3); |
| 1464 | $LD r6,`6*$BNSZ`(r4) |
| 1465 | $LD r7,`6*$BNSZ`(r5) |
| 1466 | $UMULL r8,r6,r7 |
| 1467 | $UMULH r9,r6,r7 |
| 1468 | addc r10,r10,r8 |
| 1469 | adde r11,r11,r9 |
| 1470 | addze r12,r12 |
| 1471 | #mul_add_c(a[5],b[7],c1,c2,c3); |
| 1472 | $LD r6,`5*$BNSZ`(r4) |
| 1473 | $LD r7,`7*$BNSZ`(r5) |
| 1474 | $UMULL r8,r6,r7 |
| 1475 | $UMULH r9,r6,r7 |
| 1476 | addc r10,r10,r8 |
| 1477 | adde r11,r11,r9 |
| 1478 | addze r12,r12 |
| 1479 | $ST r10,`12*$BNSZ`(r3) #r[12]=c1; |
| 1480 | #mul_add_c(a[6],b[7],c2,c3,c1); |
| 1481 | $LD r6,`6*$BNSZ`(r4) |
| 1482 | $UMULL r8,r6,r7 |
| 1483 | $UMULH r9,r6,r7 |
| 1484 | addc r11,r11,r8 |
| 1485 | adde r12,r12,r9 |
| 1486 | addze r10,r0 |
| 1487 | #mul_add_c(a[7],b[6],c2,c3,c1); |
| 1488 | $LD r6,`7*$BNSZ`(r4) |
| 1489 | $LD r7,`6*$BNSZ`(r5) |
| 1490 | $UMULL r8,r6,r7 |
| 1491 | $UMULH r9,r6,r7 |
| 1492 | addc r11,r11,r8 |
| 1493 | adde r12,r12,r9 |
| 1494 | addze r10,r10 |
| 1495 | $ST r11,`13*$BNSZ`(r3) #r[13]=c2; |
| 1496 | #mul_add_c(a[7],b[7],c3,c1,c2); |
| 1497 | $LD r7,`7*$BNSZ`(r5) |
| 1498 | $UMULL r8,r6,r7 |
| 1499 | $UMULH r9,r6,r7 |
| 1500 | addc r12,r12,r8 |
| 1501 | adde r10,r10,r9 |
| 1502 | $ST r12,`14*$BNSZ`(r3) #r[14]=c3; |
| 1503 | $ST r10,`15*$BNSZ`(r3) #r[15]=c1; |
| 1504 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1505 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1506 | |
| 1507 | # |
| 1508 | # NOTE: The following label name should be changed to |
| 1509 | # "bn_sub_words" i.e. remove the first dot |
| 1510 | # for the gcc compiler. This should be automatically |
| 1511 | # done in the build |
| 1512 | # |
| 1513 | # |
| 1514 | .align 4 |
| 1515 | .bn_sub_words: |
| 1516 | # |
| 1517 | # Handcoded version of bn_sub_words |
| 1518 | # |
| 1519 | #BN_ULONG bn_sub_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n) |
| 1520 | # |
| 1521 | # r3 = r |
| 1522 | # r4 = a |
| 1523 | # r5 = b |
| 1524 | # r6 = n |
| 1525 | # |
| 1526 | # Note: No loop unrolling done since this is not a performance |
| 1527 | # critical loop. |
| 1528 | |
| 1529 | xor r0,r0,r0 #set r0 = 0 |
| 1530 | # |
| 1531 | # check for r6 = 0 AND set carry bit. |
| 1532 | # |
| 1533 | subfc. r7,r0,r6 # If r6 is 0 then result is 0. |
| 1534 | # if r6 > 0 then result !=0 |
| 1535 | # In either case carry bit is set. |
| 1536 | beq Lppcasm_sub_adios |
| 1537 | addi r4,r4,-$BNSZ |
| 1538 | addi r3,r3,-$BNSZ |
| 1539 | addi r5,r5,-$BNSZ |
| 1540 | mtctr r6 |
| 1541 | Lppcasm_sub_mainloop: |
| 1542 | $LDU r7,$BNSZ(r4) |
| 1543 | $LDU r8,$BNSZ(r5) |
| 1544 | subfe r6,r8,r7 # r6 = r7+carry bit + onescomplement(r8) |
| 1545 | # if carry = 1 this is r7-r8. Else it |
| 1546 | # is r7-r8 -1 as we need. |
| 1547 | $STU r6,$BNSZ(r3) |
| 1548 | bdnz- Lppcasm_sub_mainloop |
| 1549 | Lppcasm_sub_adios: |
| 1550 | subfze r3,r0 # if carry bit is set then r3 = 0 else -1 |
| 1551 | andi. r3,r3,1 # keep only last bit. |
| 1552 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1553 | .long 0x00000000 |
| 1554 | |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1555 | |
| 1556 | # |
| 1557 | # NOTE: The following label name should be changed to |
| 1558 | # "bn_add_words" i.e. remove the first dot |
| 1559 | # for the gcc compiler. This should be automatically |
| 1560 | # done in the build |
| 1561 | # |
| 1562 | |
| 1563 | .align 4 |
| 1564 | .bn_add_words: |
| 1565 | # |
| 1566 | # Handcoded version of bn_add_words |
| 1567 | # |
| 1568 | #BN_ULONG bn_add_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n) |
| 1569 | # |
| 1570 | # r3 = r |
| 1571 | # r4 = a |
| 1572 | # r5 = b |
| 1573 | # r6 = n |
| 1574 | # |
| 1575 | # Note: No loop unrolling done since this is not a performance |
| 1576 | # critical loop. |
| 1577 | |
| 1578 | xor r0,r0,r0 |
| 1579 | # |
| 1580 | # check for r6 = 0. Is this needed? |
| 1581 | # |
| 1582 | addic. r6,r6,0 #test r6 and clear carry bit. |
| 1583 | beq Lppcasm_add_adios |
| 1584 | addi r4,r4,-$BNSZ |
| 1585 | addi r3,r3,-$BNSZ |
| 1586 | addi r5,r5,-$BNSZ |
| 1587 | mtctr r6 |
| 1588 | Lppcasm_add_mainloop: |
| 1589 | $LDU r7,$BNSZ(r4) |
| 1590 | $LDU r8,$BNSZ(r5) |
| 1591 | adde r8,r7,r8 |
| 1592 | $STU r8,$BNSZ(r3) |
| 1593 | bdnz- Lppcasm_add_mainloop |
| 1594 | Lppcasm_add_adios: |
| 1595 | addze r3,r0 #return carry bit. |
| 1596 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1597 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1598 | |
| 1599 | # |
| 1600 | # NOTE: The following label name should be changed to |
| 1601 | # "bn_div_words" i.e. remove the first dot |
| 1602 | # for the gcc compiler. This should be automatically |
| 1603 | # done in the build |
| 1604 | # |
| 1605 | |
| 1606 | .align 4 |
| 1607 | .bn_div_words: |
| 1608 | # |
| 1609 | # This is a cleaned up version of code generated by |
| 1610 | # the AIX compiler. The only optimization is to use |
| 1611 | # the PPC instruction to count leading zeros instead |
| 1612 | # of call to num_bits_word. Since this was compiled |
| 1613 | # only at level -O2 we can possibly squeeze it more? |
| 1614 | # |
| 1615 | # r3 = h |
| 1616 | # r4 = l |
| 1617 | # r5 = d |
| 1618 | |
| 1619 | $UCMPI 0,r5,0 # compare r5 and 0 |
| 1620 | bne Lppcasm_div1 # proceed if d!=0 |
| 1621 | li r3,-1 # d=0 return -1 |
| 1622 | blr |
| 1623 | Lppcasm_div1: |
| 1624 | xor r0,r0,r0 #r0=0 |
| 1625 | li r8,$BITS |
| 1626 | $CNTLZ. r7,r5 #r7 = num leading 0s in d. |
| 1627 | beq Lppcasm_div2 #proceed if no leading zeros |
| 1628 | subf r8,r7,r8 #r8 = BN_num_bits_word(d) |
| 1629 | $SHR. r9,r3,r8 #are there any bits above r8'th? |
| 1630 | $TR 16,r9,r0 #if there're, signal to dump core... |
| 1631 | Lppcasm_div2: |
| 1632 | $UCMP 0,r3,r5 #h>=d? |
| 1633 | blt Lppcasm_div3 #goto Lppcasm_div3 if not |
| 1634 | subf r3,r5,r3 #h-=d ; |
| 1635 | Lppcasm_div3: #r7 = BN_BITS2-i. so r7=i |
| 1636 | cmpi 0,0,r7,0 # is (i == 0)? |
| 1637 | beq Lppcasm_div4 |
| 1638 | $SHL r3,r3,r7 # h = (h<< i) |
| 1639 | $SHR r8,r4,r8 # r8 = (l >> BN_BITS2 -i) |
| 1640 | $SHL r5,r5,r7 # d<<=i |
| 1641 | or r3,r3,r8 # h = (h<<i)|(l>>(BN_BITS2-i)) |
| 1642 | $SHL r4,r4,r7 # l <<=i |
| 1643 | Lppcasm_div4: |
| 1644 | $SHRI r9,r5,`$BITS/2` # r9 = dh |
| 1645 | # dl will be computed when needed |
| 1646 | # as it saves registers. |
| 1647 | li r6,2 #r6=2 |
| 1648 | mtctr r6 #counter will be in count. |
| 1649 | Lppcasm_divouterloop: |
| 1650 | $SHRI r8,r3,`$BITS/2` #r8 = (h>>BN_BITS4) |
| 1651 | $SHRI r11,r4,`$BITS/2` #r11= (l&BN_MASK2h)>>BN_BITS4 |
| 1652 | # compute here for innerloop. |
| 1653 | $UCMP 0,r8,r9 # is (h>>BN_BITS4)==dh |
| 1654 | bne Lppcasm_div5 # goto Lppcasm_div5 if not |
| 1655 | |
| 1656 | li r8,-1 |
| 1657 | $CLRU r8,r8,`$BITS/2` #q = BN_MASK2l |
| 1658 | b Lppcasm_div6 |
| 1659 | Lppcasm_div5: |
| 1660 | $UDIV r8,r3,r9 #q = h/dh |
| 1661 | Lppcasm_div6: |
| 1662 | $UMULL r12,r9,r8 #th = q*dh |
| 1663 | $CLRU r10,r5,`$BITS/2` #r10=dl |
| 1664 | $UMULL r6,r8,r10 #tl = q*dl |
| 1665 | |
| 1666 | Lppcasm_divinnerloop: |
| 1667 | subf r10,r12,r3 #t = h -th |
| 1668 | $SHRI r7,r10,`$BITS/2` #r7= (t &BN_MASK2H), sort of... |
| 1669 | addic. r7,r7,0 #test if r7 == 0. used below. |
| 1670 | # now want to compute |
| 1671 | # r7 = (t<<BN_BITS4)|((l&BN_MASK2h)>>BN_BITS4) |
| 1672 | # the following 2 instructions do that |
| 1673 | $SHLI r7,r10,`$BITS/2` # r7 = (t<<BN_BITS4) |
| 1674 | or r7,r7,r11 # r7|=((l&BN_MASK2h)>>BN_BITS4) |
| 1675 | $UCMP cr1,r6,r7 # compare (tl <= r7) |
| 1676 | bne Lppcasm_divinnerexit |
| 1677 | ble cr1,Lppcasm_divinnerexit |
| 1678 | addi r8,r8,-1 #q-- |
| 1679 | subf r12,r9,r12 #th -=dh |
| 1680 | $CLRU r10,r5,`$BITS/2` #r10=dl. t is no longer needed in loop. |
| 1681 | subf r6,r10,r6 #tl -=dl |
| 1682 | b Lppcasm_divinnerloop |
| 1683 | Lppcasm_divinnerexit: |
| 1684 | $SHRI r10,r6,`$BITS/2` #t=(tl>>BN_BITS4) |
| 1685 | $SHLI r11,r6,`$BITS/2` #tl=(tl<<BN_BITS4)&BN_MASK2h; |
| 1686 | $UCMP cr1,r4,r11 # compare l and tl |
| 1687 | add r12,r12,r10 # th+=t |
| 1688 | bge cr1,Lppcasm_div7 # if (l>=tl) goto Lppcasm_div7 |
| 1689 | addi r12,r12,1 # th++ |
| 1690 | Lppcasm_div7: |
| 1691 | subf r11,r11,r4 #r11=l-tl |
| 1692 | $UCMP cr1,r3,r12 #compare h and th |
| 1693 | bge cr1,Lppcasm_div8 #if (h>=th) goto Lppcasm_div8 |
| 1694 | addi r8,r8,-1 # q-- |
| 1695 | add r3,r5,r3 # h+=d |
| 1696 | Lppcasm_div8: |
| 1697 | subf r12,r12,r3 #r12 = h-th |
| 1698 | $SHLI r4,r11,`$BITS/2` #l=(l&BN_MASK2l)<<BN_BITS4 |
| 1699 | # want to compute |
| 1700 | # h = ((h<<BN_BITS4)|(l>>BN_BITS4))&BN_MASK2 |
| 1701 | # the following 2 instructions will do this. |
| 1702 | $INSR r11,r12,`$BITS/2`,`$BITS/2` # r11 is the value we want rotated $BITS/2. |
| 1703 | $ROTL r3,r11,`$BITS/2` # rotate by $BITS/2 and store in r3 |
| 1704 | bdz Lppcasm_div9 #if (count==0) break ; |
| 1705 | $SHLI r0,r8,`$BITS/2` #ret =q<<BN_BITS4 |
| 1706 | b Lppcasm_divouterloop |
| 1707 | Lppcasm_div9: |
| 1708 | or r3,r8,r0 |
| 1709 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1710 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1711 | |
| 1712 | # |
| 1713 | # NOTE: The following label name should be changed to |
| 1714 | # "bn_sqr_words" i.e. remove the first dot |
| 1715 | # for the gcc compiler. This should be automatically |
| 1716 | # done in the build |
| 1717 | # |
| 1718 | .align 4 |
| 1719 | .bn_sqr_words: |
| 1720 | # |
| 1721 | # Optimized version of bn_sqr_words |
| 1722 | # |
| 1723 | # void bn_sqr_words(BN_ULONG *r, BN_ULONG *a, int n) |
| 1724 | # |
| 1725 | # r3 = r |
| 1726 | # r4 = a |
| 1727 | # r5 = n |
| 1728 | # |
| 1729 | # r6 = a[i]. |
| 1730 | # r7,r8 = product. |
| 1731 | # |
| 1732 | # No unrolling done here. Not performance critical. |
| 1733 | |
| 1734 | addic. r5,r5,0 #test r5. |
| 1735 | beq Lppcasm_sqr_adios |
| 1736 | addi r4,r4,-$BNSZ |
| 1737 | addi r3,r3,-$BNSZ |
| 1738 | mtctr r5 |
| 1739 | Lppcasm_sqr_mainloop: |
| 1740 | #sqr(r[0],r[1],a[0]); |
| 1741 | $LDU r6,$BNSZ(r4) |
| 1742 | $UMULL r7,r6,r6 |
| 1743 | $UMULH r8,r6,r6 |
| 1744 | $STU r7,$BNSZ(r3) |
| 1745 | $STU r8,$BNSZ(r3) |
| 1746 | bdnz- Lppcasm_sqr_mainloop |
| 1747 | Lppcasm_sqr_adios: |
| 1748 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1749 | .long 0x00000000 |
| 1750 | |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1751 | |
| 1752 | # |
| 1753 | # NOTE: The following label name should be changed to |
| 1754 | # "bn_mul_words" i.e. remove the first dot |
| 1755 | # for the gcc compiler. This should be automatically |
| 1756 | # done in the build |
| 1757 | # |
| 1758 | |
| 1759 | .align 4 |
| 1760 | .bn_mul_words: |
| 1761 | # |
| 1762 | # BN_ULONG bn_mul_words(BN_ULONG *rp, BN_ULONG *ap, int num, BN_ULONG w) |
| 1763 | # |
| 1764 | # r3 = rp |
| 1765 | # r4 = ap |
| 1766 | # r5 = num |
| 1767 | # r6 = w |
| 1768 | xor r0,r0,r0 |
| 1769 | xor r12,r12,r12 # used for carry |
| 1770 | rlwinm. r7,r5,30,2,31 # num >> 2 |
| 1771 | beq Lppcasm_mw_REM |
| 1772 | mtctr r7 |
| 1773 | Lppcasm_mw_LOOP: |
| 1774 | #mul(rp[0],ap[0],w,c1); |
| 1775 | $LD r8,`0*$BNSZ`(r4) |
| 1776 | $UMULL r9,r6,r8 |
| 1777 | $UMULH r10,r6,r8 |
| 1778 | addc r9,r9,r12 |
| 1779 | #addze r10,r10 #carry is NOT ignored. |
| 1780 | #will be taken care of |
| 1781 | #in second spin below |
| 1782 | #using adde. |
| 1783 | $ST r9,`0*$BNSZ`(r3) |
| 1784 | #mul(rp[1],ap[1],w,c1); |
| 1785 | $LD r8,`1*$BNSZ`(r4) |
| 1786 | $UMULL r11,r6,r8 |
| 1787 | $UMULH r12,r6,r8 |
| 1788 | adde r11,r11,r10 |
| 1789 | #addze r12,r12 |
| 1790 | $ST r11,`1*$BNSZ`(r3) |
| 1791 | #mul(rp[2],ap[2],w,c1); |
| 1792 | $LD r8,`2*$BNSZ`(r4) |
| 1793 | $UMULL r9,r6,r8 |
| 1794 | $UMULH r10,r6,r8 |
| 1795 | adde r9,r9,r12 |
| 1796 | #addze r10,r10 |
| 1797 | $ST r9,`2*$BNSZ`(r3) |
| 1798 | #mul_add(rp[3],ap[3],w,c1); |
| 1799 | $LD r8,`3*$BNSZ`(r4) |
| 1800 | $UMULL r11,r6,r8 |
| 1801 | $UMULH r12,r6,r8 |
| 1802 | adde r11,r11,r10 |
| 1803 | addze r12,r12 #this spin we collect carry into |
| 1804 | #r12 |
| 1805 | $ST r11,`3*$BNSZ`(r3) |
| 1806 | |
| 1807 | addi r3,r3,`4*$BNSZ` |
| 1808 | addi r4,r4,`4*$BNSZ` |
| 1809 | bdnz- Lppcasm_mw_LOOP |
| 1810 | |
| 1811 | Lppcasm_mw_REM: |
| 1812 | andi. r5,r5,0x3 |
| 1813 | beq Lppcasm_mw_OVER |
| 1814 | #mul(rp[0],ap[0],w,c1); |
| 1815 | $LD r8,`0*$BNSZ`(r4) |
| 1816 | $UMULL r9,r6,r8 |
| 1817 | $UMULH r10,r6,r8 |
| 1818 | addc r9,r9,r12 |
| 1819 | addze r10,r10 |
| 1820 | $ST r9,`0*$BNSZ`(r3) |
| 1821 | addi r12,r10,0 |
| 1822 | |
| 1823 | addi r5,r5,-1 |
| 1824 | cmpli 0,0,r5,0 |
| 1825 | beq Lppcasm_mw_OVER |
| 1826 | |
| 1827 | |
| 1828 | #mul(rp[1],ap[1],w,c1); |
| 1829 | $LD r8,`1*$BNSZ`(r4) |
| 1830 | $UMULL r9,r6,r8 |
| 1831 | $UMULH r10,r6,r8 |
| 1832 | addc r9,r9,r12 |
| 1833 | addze r10,r10 |
| 1834 | $ST r9,`1*$BNSZ`(r3) |
| 1835 | addi r12,r10,0 |
| 1836 | |
| 1837 | addi r5,r5,-1 |
| 1838 | cmpli 0,0,r5,0 |
| 1839 | beq Lppcasm_mw_OVER |
| 1840 | |
| 1841 | #mul_add(rp[2],ap[2],w,c1); |
| 1842 | $LD r8,`2*$BNSZ`(r4) |
| 1843 | $UMULL r9,r6,r8 |
| 1844 | $UMULH r10,r6,r8 |
| 1845 | addc r9,r9,r12 |
| 1846 | addze r10,r10 |
| 1847 | $ST r9,`2*$BNSZ`(r3) |
| 1848 | addi r12,r10,0 |
| 1849 | |
| 1850 | Lppcasm_mw_OVER: |
| 1851 | addi r3,r12,0 |
| 1852 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1853 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1854 | |
| 1855 | # |
| 1856 | # NOTE: The following label name should be changed to |
| 1857 | # "bn_mul_add_words" i.e. remove the first dot |
| 1858 | # for the gcc compiler. This should be automatically |
| 1859 | # done in the build |
| 1860 | # |
| 1861 | |
| 1862 | .align 4 |
| 1863 | .bn_mul_add_words: |
| 1864 | # |
| 1865 | # BN_ULONG bn_mul_add_words(BN_ULONG *rp, BN_ULONG *ap, int num, BN_ULONG w) |
| 1866 | # |
| 1867 | # r3 = rp |
| 1868 | # r4 = ap |
| 1869 | # r5 = num |
| 1870 | # r6 = w |
| 1871 | # |
| 1872 | # empirical evidence suggests that unrolled version performs best!! |
| 1873 | # |
| 1874 | xor r0,r0,r0 #r0 = 0 |
| 1875 | xor r12,r12,r12 #r12 = 0 . used for carry |
| 1876 | rlwinm. r7,r5,30,2,31 # num >> 2 |
| 1877 | beq Lppcasm_maw_leftover # if (num < 4) go LPPCASM_maw_leftover |
| 1878 | mtctr r7 |
| 1879 | Lppcasm_maw_mainloop: |
| 1880 | #mul_add(rp[0],ap[0],w,c1); |
| 1881 | $LD r8,`0*$BNSZ`(r4) |
| 1882 | $LD r11,`0*$BNSZ`(r3) |
| 1883 | $UMULL r9,r6,r8 |
| 1884 | $UMULH r10,r6,r8 |
| 1885 | addc r9,r9,r12 #r12 is carry. |
| 1886 | addze r10,r10 |
| 1887 | addc r9,r9,r11 |
| 1888 | #addze r10,r10 |
| 1889 | #the above instruction addze |
| 1890 | #is NOT needed. Carry will NOT |
| 1891 | #be ignored. It's not affected |
| 1892 | #by multiply and will be collected |
| 1893 | #in the next spin |
| 1894 | $ST r9,`0*$BNSZ`(r3) |
| 1895 | |
| 1896 | #mul_add(rp[1],ap[1],w,c1); |
| 1897 | $LD r8,`1*$BNSZ`(r4) |
| 1898 | $LD r9,`1*$BNSZ`(r3) |
| 1899 | $UMULL r11,r6,r8 |
| 1900 | $UMULH r12,r6,r8 |
| 1901 | adde r11,r11,r10 #r10 is carry. |
| 1902 | addze r12,r12 |
| 1903 | addc r11,r11,r9 |
| 1904 | #addze r12,r12 |
| 1905 | $ST r11,`1*$BNSZ`(r3) |
| 1906 | |
| 1907 | #mul_add(rp[2],ap[2],w,c1); |
| 1908 | $LD r8,`2*$BNSZ`(r4) |
| 1909 | $UMULL r9,r6,r8 |
| 1910 | $LD r11,`2*$BNSZ`(r3) |
| 1911 | $UMULH r10,r6,r8 |
| 1912 | adde r9,r9,r12 |
| 1913 | addze r10,r10 |
| 1914 | addc r9,r9,r11 |
| 1915 | #addze r10,r10 |
| 1916 | $ST r9,`2*$BNSZ`(r3) |
| 1917 | |
| 1918 | #mul_add(rp[3],ap[3],w,c1); |
| 1919 | $LD r8,`3*$BNSZ`(r4) |
| 1920 | $UMULL r11,r6,r8 |
| 1921 | $LD r9,`3*$BNSZ`(r3) |
| 1922 | $UMULH r12,r6,r8 |
| 1923 | adde r11,r11,r10 |
| 1924 | addze r12,r12 |
| 1925 | addc r11,r11,r9 |
| 1926 | addze r12,r12 |
| 1927 | $ST r11,`3*$BNSZ`(r3) |
| 1928 | addi r3,r3,`4*$BNSZ` |
| 1929 | addi r4,r4,`4*$BNSZ` |
| 1930 | bdnz- Lppcasm_maw_mainloop |
| 1931 | |
| 1932 | Lppcasm_maw_leftover: |
| 1933 | andi. r5,r5,0x3 |
| 1934 | beq Lppcasm_maw_adios |
| 1935 | addi r3,r3,-$BNSZ |
| 1936 | addi r4,r4,-$BNSZ |
| 1937 | #mul_add(rp[0],ap[0],w,c1); |
| 1938 | mtctr r5 |
| 1939 | $LDU r8,$BNSZ(r4) |
| 1940 | $UMULL r9,r6,r8 |
| 1941 | $UMULH r10,r6,r8 |
| 1942 | $LDU r11,$BNSZ(r3) |
| 1943 | addc r9,r9,r11 |
| 1944 | addze r10,r10 |
| 1945 | addc r9,r9,r12 |
| 1946 | addze r12,r10 |
| 1947 | $ST r9,0(r3) |
| 1948 | |
| 1949 | bdz Lppcasm_maw_adios |
| 1950 | #mul_add(rp[1],ap[1],w,c1); |
| 1951 | $LDU r8,$BNSZ(r4) |
| 1952 | $UMULL r9,r6,r8 |
| 1953 | $UMULH r10,r6,r8 |
| 1954 | $LDU r11,$BNSZ(r3) |
| 1955 | addc r9,r9,r11 |
| 1956 | addze r10,r10 |
| 1957 | addc r9,r9,r12 |
| 1958 | addze r12,r10 |
| 1959 | $ST r9,0(r3) |
| 1960 | |
| 1961 | bdz Lppcasm_maw_adios |
| 1962 | #mul_add(rp[2],ap[2],w,c1); |
| 1963 | $LDU r8,$BNSZ(r4) |
| 1964 | $UMULL r9,r6,r8 |
| 1965 | $UMULH r10,r6,r8 |
| 1966 | $LDU r11,$BNSZ(r3) |
| 1967 | addc r9,r9,r11 |
| 1968 | addze r10,r10 |
| 1969 | addc r9,r9,r12 |
| 1970 | addze r12,r10 |
| 1971 | $ST r9,0(r3) |
| 1972 | |
| 1973 | Lppcasm_maw_adios: |
| 1974 | addi r3,r12,0 |
| 1975 | blr |
Alexandre Savard | 7541067 | 2012-08-08 09:50:01 -0400 | [diff] [blame] | 1976 | .long 0x00000000 |
Alexandre Savard | 1b09e31 | 2012-08-07 20:33:29 -0400 | [diff] [blame] | 1977 | .align 4 |
| 1978 | EOF |
| 1979 | $data =~ s/\`([^\`]*)\`/eval $1/gem; |
| 1980 | print $data; |
| 1981 | close STDOUT; |